Publications of Paris Mastorocostas

 

Last update: October 3, 2024

 

     This is the complete list of my publications (accepted, published). To get a hard copy of any of these papers please e-mail me at mast@ieee.org and you will receive it either by surface or electronic mail.


 

Journal Papers


 

Optimizing UAV-Based Inventory Detection and Quantification in Industrial WarehousesL A LiDAR-Driven Approach

 

S. Tsakiridis, A. Papakonstantinou, A. Kapandelis, P. Mastorocostas, A. Tsimpiris, D. Varsamis

WSEAS Transactions on Systems, vol. 23, pp. 121-127, 2024

 

Abstract

The advancement of technology has brought about a revolution in industrial operations, where specialized tools play a crucial role in enhancing efficiency. This study delves into the significant impact of the logistics department in global industries and proposes an innovative solution for inventory detection and recognition using unmanned aerial vehicles (UAVs) equipped with LiDAR technology. Unlike existing research that often involves intricate hardware systems and algorithms leading to increased costs and computational demands, our research focuses on streamlining the inventory detection process by utilizing a LiDAR data and an algorithmic approach that minimizes the time of extensive counting process into the warehouse to quantify the pallets existing. The proposed methodology entails a custom-made quadcopter equipped with a single-beam and high-frequency LiDAR range finder. Operating autonomously along a predetermined flight plan, the drone captures high-frequency range data of warehouse inventory. The paper comprehensively outlines the UAV control procedures, warehouse scanning using LiDAR, and the inventory detection and quantification of pallets algorithmic process. The proposed method processes LiDAR data in a post-process way, estimating the number of pallets and, consequently, producing a map of each stack within the warehouse denoting the quantities of pallets. The research results showcase the successful implementation of the proposed approach in a model warehouse, achieving an impressive 100% evaluation accuracy. Future research endeavors aim to extend this methodology to warehouses with dynamic product placements, emphasizing real-time monitoring for comprehensive inventory detection. This innovative approach stands out as a cost-effective and efficient solution for industries seeking accurate and timely inventory information.


 

Computational Techniques for Locating Industrial Products in Warehouses

 

S. Tsakiridis, A. Papakonstantinou, A. Kapandelis, P. Mastorocostas, A. Tsimpiris, D. Varsamis

Contemporary Engineering Sciences, vol. 16, no 1, pp. 71-79, 2023

 

Abstract

The computational estimation of an indoor or open-space warehouse inventory is based on the prior knowledge of the pallet dimensions. The input data consist of a three-dimensional point-cloud created by three-dimensional (3D) scanners (LiDAR technology) adapted to aerial vehicles (Drones). For research purposes, a storage simulator has been implemented in the Python language (version 3.9). In the _rst phase, this research focuses on the ideal case of point-dispersion, with integer values for coordinates, in which a unit length corresponds to the distance between two neighboring pixels in the horizontal or vertical direction. In a subsequent stage, the generator of three-dimensional points will be modifed to produce more realistic warehouse models. Improved versions of existing algorithms will be proposed, taking into consideration the height variations.


 

A Spatial-Temporal Graph Convolutional Recurrent Network for Transportation Flow Estimation

 

I. Drosouli, A. Voulodimos, P. Mastorocostas, G. Miaoulis,  D. Ghazanfarpour

Sensors, vol. 23, no 7534, pp. 1-25, 2023

 

Abstract

Accurate estimation of transportation flow is a challenging task in Intelligent Transportation Systems (ITS). Transport data with dynamic spatial-temporal dependencies elevates transportation flow forecasting to a significant issue for operational planning, managing passenger flow, and arranging for individual travel in a smart city. The task is challenging due to the composite spatial dependency on transportation networks and the non-linear temporal dynamics with mobility conditions changing over time. To address these challenges, we propose ST-GCRN, a Spatial-Temporal Graph Convolutional Recurrent Network that learns from both the spatial stations network data and time-series of historical mobility changes so as to estimate transportation flow at a future time. The model is based on Graph Convolutional Networks (GCN) and Long Short-Term Memory (LSTM) in order to further improve the accuracy of transportation flow estimation. Extensive experiments on two real-world datasets of transportation flow, New York bike-sharing system and Hangzhou metro system, prove the effectiveness of the proposed model.


 

Short-Term Load Forecasting of the Greek Power System Using a Dynamic Block-Diagonal Fuzzy Neural Network

 

G, Kandilogiannakis, P. Mastorocostas, A. Voulodimos, C. Hilas

Energies, vol. 16, no 4227, pp. 1-21, 2023

 

Abstract

A dynamic fuzzy neural network for short-term load forecasting of the Greek power system is proposed, and hourly-based prediction for the whole year is performed. DBD-FELF (Dynamic Block-Diagonal Fuzzy Electric Load Forecaster) consists of fuzzy rules with consequent parts that are neural networks with internal recurrence. These networks have a hidden layer which consists of pairs of neurons with feedback connections between them. The overall fuzzy model partitions the input space in partially overlapping fuzzy regions, where the recurrent neural networks of the respective rules operate. The partition of the input space and determination of the fuzzy rule base is performed by use of Fuzzy C-Means clustering algorithm and the RENNCOM constrained optimization method is applied for consequent parameter tuning. The performance of DBD-FELF is tested via extensive experimental analysis and the results are promising, since an average percentage error of 1.18% is attained, along with an average yearly absolute error of 76.2 MW. Moreover, DBD-FELF is compared with Deep Learning, fuzzy and neurofuzzy rivals, such that its particular attributes are highlighted.


 

TDM-BERT: A Transformer-Based Model for Transportation Mode Detection

 

I. Drosouli, A. Voulodimos, P. Mastorocostas, G. Miaoulis,  D. Ghazanfarpour

Electronics, vol. 12, no 581, pp. 1-17, 2023

 

Abstract

Aiming to differentiate various transportation modes and detect the means of transport an individual uses, is the focal point of transportation mode detection, one of the problems in the field of intelligent transport which receives the attention of researchers because of its interesting and useful applications. In this paper we present TMD-BERT, a transformer based model for transportation mode detection based on sensor data. The proposed transformer-based approach processes the entire sequence of data, understand the importance of each part of the input sequence and assign weights accordingly, using attention mechanisms, to learn global dependencies in the sequence. The experimental evaluation shows the high performance of the model compared to the state of the art, demonstrating a prediction accuracy of 98.8%.



 

TDM-BERT: A Transformer-Based Model for Transportation Mode Detection

 

I. Drosouli, A. Voulodimos, P. Mastorocostas, G. Miaoulis,  D. Ghazanfarpour

Electronics, vol. 12, no 581, pp. 1-17, 2023

 

Abstract

Aiming to differentiate various transportation modes and detect the means of transport an individual uses, is the focal point of transportation mode detection, one of the problems in the field of intelligent transport which receives the attention of researchers because of its interesting and useful applications. In this paper we present TMD-BERT, a transformer based model for transportation mode detection based on sensor data. The proposed transformer-based approach processes the entire sequence of data, understand the importance of each part of the input sequence and assign weights accordingly, using attention mechanisms, to learn global dependencies in the sequence. The experimental evaluation shows the high performance of the model compared to the state of the art, demonstrating a prediction accuracy of 98.8%.


 

Computational Techniques for Locating Industrial Products in Warehouses

 

S. Tsakiridis, A. Papakonstantinou, A. Kapandelis, P. Mastorocostas, A. Tsimpiris, D. Varsamis, “Computational Techniques for Locating Industrial Products in Warehouses,” Contemporary Engineering Sciences, vol. 16, no 1, pp. 71-79, 2023.

 

Abstract

The computational estimation of an indoor or open-space warehouse inventory is based on the prior knowledge of the pallet dimensions. The input data consist of a three-dimensional point-cloud created by three-dimensional (3D) scanners (LiDAR technology) adapted to aerial vehicles (Drones). For research purposes, a storage simulator has been implemented in the Python language (version 3.9). In the _rst phase, this research focuses on the ideal case of point-dispersion, with integer values for coordinates, in which a unit length corresponds to the distance between two neighboring pixels in the horizontal or vertical direction. In a subsequent stage, the generator of three-dimensional points will be modifed to produce more realistic warehouse models. Improved versions of existing algorithms will be proposed, taking into consideration the height variations.


 

ReNFuzz-LF: A Recurrent Neurofuzzy Model for Short-Term Load Forecasting

 

G, Kandilogiannakis, P. Mastorocostas, A. Voulodimos

Energies, vol. 15, no 3637, pp. 1-18, 2022

 

Abstract

A neurofuzzy system is proposed for short-term electric load forecasting. The fuzzy rule base of ReNFuzz-LF consists of rules with dynamic consequent parts that are small-scale recurrent neural networks with one hidden layer, whose neurons have local output feedback. The particular representation maintains the local learning nature of the typical static fuzzy model, since the dynamic consequent parts of the fuzzy rules can be considered as subsystems operating at the subspaces defined by the fuzzy premise parts, and they are interconnected through the defuzzification part. The Greek power system is examined, and hourly based predictions are extracted for the whole year. The recurrent nature of the forecaster leads to the use of a minimal set of inputs, since the temporal relations of the electric load time-series are identified without any prior knowledge of the appropriate past load values being necessary. An extensive simulation analysis is conducted, and the forecaster’s performance is evaluated using appropriate metrics (APE, RMSE, forecast error duration curve). ReNFuzz-LF performs efficiently, attaining an average percentage error of 1.35% and an average yearly absolute error of 86.3 MW. Finally, the performance of the proposed forecaster is compared to a series of Computational Intelligence based models, such that the learning characteristics of ReNFuzz-LF are highlighted.


 

Transportation Mode Detection Using an Optimized Long Short-Term Model on Multimodal Sensor Data

 

I. Drosouli, A. Voulodimos, G. Miaoulis, P. Mastorocostas, D. Ghazanfarpour

Entropy, vol. 23, no 1457, pp. 1-20, 2021

 

Abstract

The advancement of sensing technologies coupled with the rapid progress in big data analysis has ushered a new era in intelligent transport and smart city applications. In this context, transportation mode detection (TMD) of mobile users is a field that has gained significant traction in recent years. In this paper, we present a deep learning approach for transportation mode detection using multimodal sensor data elicited from user smartphone. The approach is based on Long Short-Term Memory networks and Bayesian optimization of their parameters. We have conducted an extensive experimental evaluation of the proposed approach, which attains very high recognition rates, against a multitude of machine learning approaches, including state of the art methods, and also discuss issues regarding feature correlation and impact of dimensionality reduction.


 

Neurofuzzy Modelling of Lung Sounds

 

G. Kandilogiannakis, P. Mastorocostas, D. Varsamis, C. Hilas,

Contemporary Engineering Sciences, vol. 11, no 98, pp. 4879-4890, 2018

 

Abstract

In this paper a computational intelligence-based filter for real-time separation the adventitious discontinuous lung sounds from the vesicular sounds is proposed. The filter uses two Dynamic Fuzzy Neural Networks to perform the task of separation of the lung sounds, obtained from patients with pulmonary pathology. The networks are trained by the Simulated Annealing Dynamic Resilient Propagation algorithm and the resulting filter is applied to three major classes of lung sounds. In order to highlight the learning characteristics and the performance of the proposed separation scheme, extensive experimental analysis is conducted, where a comparison with other filters is given.


 

A Hybrid Learning Algorithm for Locally Recurrent Neural Networks

 

D. Varsamis, E. Outsios, P. Mastorocostas,

Contemporary Engineering Sciences, vol. 11, no 1, pp. 1-13, 2018

 

Abstract

In this work a fast and efficient training method for block-diagonal recurrent neural networks is proposed. The method modifies and extends the Simulated Annealing RPROP algorithm, originally developed for static models, by taking into consideration the architectural characteristics and the temporal nature of this category of recurrent neural models. The performance of the proposed algorithm is evaluated through a comparative analysis with a series of algorithms and recurrent models.


 

3D Geospatial Visualizations: Animation and Motion Effects on Spatial Objects

 

K. Evangelidis, T. Papadopoulos, K. Papatheodorou, P. Mastorocostas, C. Hilas

Computers and Geosciences, vol. 111, pp. 200-212, 2018

 

Abstract

Digital Elevation Models (DEMs), in combination with high quality raster graphics provide realistic three-dimensional (3D) representations of the globe (virtual globe) and amazing navigation experience over the terrain through earth browsers. In addition, the adoption of interoperable geospatial mark-up languages (e.g. KML) and open programming libraries (Javascript) makes it also possible to create 3D spatial objects and convey on them the sensation of any type of texture by utilizing open 3D representation models (e.g. Collada). One step beyond, by employing WebGL frameworks (e.g. Cesium.js, three.js) animation and motion effects are attributed on 3D models. However, major GIS-based functionalities in combination with all the above mentioned visualization capabilities such as for example animation effects on selected areas of the terrain texture (e.g. sea waves) as well as motion effects on 3D objects moving in dynamically defined georeferenced terrain paths (e.g. the motion of an animal over a hill, or of a big fish in an ocean etc.) are not widely supported at least by open geospatial applications or development frameworks. Towards this we developed and made available to the research community, an open geospatial software application prototype that provides high level capabilities for dynamically creating user defined virtual geospatial worlds populated by selected animated and moving 3D models on user specified locations, paths and areas. At the same time, the generated code may enhance existing open visualization frameworks and programming libraries dealing with 3D simulations, with the geospatial aspect of a virtual world.


 

Telecommunications Call Volume Forecasting with a Block-Diagonal Recurrent Fuzzy Neural Network

 

P. Mastorocostas, C. Hilas, D. Varsamis, S. Dova

Telecommunication Systems, vol. 63, no 1, pp. 15-25, 2016

 

Abstract

An application of computational intelligence to the problem of telecommunications call volume forecasting is proposed in this work. In particular, the forecasting system is a recurrent fuzzy-neural model. The premise and defuzzification parts of the model’s fuzzy rules are static, while the consequent parts of the fuzzy rules are small blockdiagonal recurrent neural networks with internal feedback, thus enabling the overall system to discover the temporal dependencies of the telecommunications time-series and perform forecasting without requiring prior knowledge of the exact order of the time-series. The forecasting performance is evaluated by using real-world telecommunications data. An extensive comparative analysis with a series of existing forecasters is conducted, including both traditional models as well as computational intelligence’s approaches. The simulation results confirm the modelling potential of the proposed scheme, since the latter outperforms its competing rivals in terms of three appropriate metrics, in all kinds of calls.


 

Transformations Between Two-Variable Polynomial Bases with Applications

 

D. Varsamis, N. Karampetakis, P. Mastorocostas

Applied Mathematics & Information Sciences, vol. 10, no 4, pp. 1303-1311, 2016

 

Abstract

This paper proposes an object-based segmentation/classification scheme for remotely sensed images, based on a novel variant of the recently proposed Genetic Sequential Image Segmentation (GeneSIS) algorithm. GeneSIS segments the image in an iterative manner, whereby at each iteration a single object is extracted via a genetic-based object extraction algorithm. In the previous version of GeneSIS, the candidate objects to be extracted were evaluated through the fuzzy content of their included pixels. In the newly developed region-based GeneSIS algorithm, a watershed-driven fine segmentation map is initially obtained from the original image, which serves as the basis for the forthcoming GeneSIS segmentation. Furthermore, in order to enhance the spatial search capabilities, we introduce a more descriptive encoding scheme in the object extraction algorithm, where the structural search modules are represented by polygonal shapes. Our objectives in the new framework are posed as follows: enhance the flexibility of the algorithm in extracting more flexible object shapes, assure high level classification accuracies, reduce the execution time of the segmentation, while at the same time preserving all the inherent attributes of the GeneSIS approach. Finally, exploiting the inherent attribute of GeneSIS to produce multiple segmentations, we also propose two segmentation fusion schemes that operate on the ensemble of segmentations generated by GeneSIS. The validity of the suggested methods is demonstrated on the land cover classification of two remotely sensed images.


 

Clustering of Telecommunications Users Profiles for Fraud Detection and Security Enhancement in Large Corporate Networks: A Case-Study

 

C. Hilas, P. Mastorocostas, I. Rekanos

Applied Mathematics & Information Sciences, vol. 9, no 4, pp. 1709-1718, 2015

 

Abstract

A user’s transactions with modern networks and services produce a vast amount of user related data. The byproduct of every phone call a person makes or every web page one visits is translated into a log record with usage data. By studying these log records, the user’s behavior is revealed and one may come up with clues about user preferences, identify security issues, or discover fraudulent use of the network or the service one provides. Thus, the modeling of network users’ behavior may serve as an invaluable tool for the IT manager. In this paper, many of these issues are discussed and emphasis is given on the construction of appropriate user behavior representation in telecommunications. As an example, the application of two clustering techniques is presented, with the task to identify appropriate user behavior representations (profiles) inside a large organization’s telecommunications network, in order to spot fraudulent usage. Through this study a researcher and/or the organization’s network manager may gain more insight into the problems of user profiling and fraud detection.


 

Classification of Remotely Sensed Images Using the GeneSIS Segmentation Algorithm for Classification of

 

S. Mylonas, D. Stavrakoudis, J. Theocharis, P. Mastorocostas

IEEE Transactions on Geoscience and Remote Sensing, vol. 53, nο 10, pp. 5352-5376, October 2015

 

Abstract

In this paper we propose an integrated framework of the recently proposed Genetic Sequential Image Segmentation (GeneSIS) algorithm. GeneSIS segments the image in an iterative manner, whereby at each iteration a single object is extracted via a genetic algorithm-based object extraction method. This module evaluates the fuzzy content of candidate regions, and through an effective fitness function design provides objects with optimal balance between fuzzy coverage, consistency and smoothness. GeneSIS exhibits a number of interesting properties, such as reduced over/under-segmentation, adaptive search scale, and region-based search. To enhance the capabilities of GeneSIS, we incorporate here several improvements of our initial proposal. On one hand, two modifications are introduced pertaining to the object extraction algorithm. Specifically, we consider a more flexible representation of the structural elements used for object’s extraction. Further, in view of its importance, the consistency criterion is redefined, thus providing a better handling of the ambiguous areas of the image. On the other hand we incorporate three tools properly devised according to the fuzzy principles characterizing GeneSIS. First, we develop a marker selection strategy that creates reliable markers, especially when dealing with ambiguous components of the image. Further, using GeneSIS as the essential part, we consider a generalized experimental setup embracing two different classification schemes for remote sensing images: the spectral-spatial classification and the supervised segmentation methods. Finally, exploiting the inherent property of GeneSIS to produce multiple segmentations, we propose a segmentation fusion scheme. The effectiveness of the proposed methodology is validated after thorough experimentation on four datasets.


 

A Region-based GeneSIS Segmentation Algorithm for Classification of Remotely Sensed Images

 

S. Mylonas, D. Stavrakoudis, J. Theocharis, P. Mastorocostas

Remote Sensing, vol. 7, pp. 2474-2508, 2015

 

Abstract

This paper proposes an object-based segmentation/classification scheme for remotely sensed images, based on a novel variant of the recently proposed Genetic Sequential Image Segmentation (GeneSIS) algorithm. GeneSIS segments the image in an iterative manner, whereby at each iteration a single object is extracted via a genetic-based object extraction algorithm. In the previous version of GeneSIS, the candidate objects to be extracted were evaluated through the fuzzy content of their included pixels. In the newly developed region-based GeneSIS algorithm, a watershed-driven fine segmentation map is initially obtained from the original image, which serves as the basis for the forthcoming GeneSIS segmentation. Furthermore, in order to enhance the spatial search capabilities, we introduce a more descriptive encoding scheme in the object extraction algorithm, where the structural search modules are represented by polygonal shapes. Our objectives in the new framework are posed as follows: enhance the flexibility of the algorithm in extracting more flexible object shapes, assure high level classification accuracies, reduce the execution time of the segmentation, while at the same time preserving all the inherent attributes of the GeneSIS approach. Finally, exploiting the inherent attribute of GeneSIS to produce multiple segmentations, we also propose two segmentation fusion schemes that operate on the ensemble of segmentations generated by GeneSIS. The validity of the suggested methods is demonstrated on the land cover classification of two remotely sensed images.


 

An Optimal Bivariate Polynomial Interpolation Basis for the Application of the Evaluation-Interpolation Technique

 

D. Varsamis, N. Karampetakis, P. Mastorocostas

Applied Mathematics & Information Sciences, vol. 8, no 1, pp. 117-125, 2014

 

Abstract

A new basis of interpolation points for the special case of the Newton two variable polynomial interpolation problem is proposed. This basis is implemented when the upper bound of the total degree and the degree in each variable is known. It is shown that this new basis under certain conditions (that depends on the degrees of the interpolation polynomial), coincides either with the known triangular/rectangular basis or it is a polygonal basis. In all cases it uses the least interpolation points with further consequences to the complexity of the algorithms that we use.


 

A Recurrent Neural Network-based Forecasting System for Telecommunications Call Volume

 

C. Hilas, I. Rekanos, P. Mastorocostas

Mathematical Problems in Engineering, doi: 10.1155/2013/317613, 2013

 

Abstract

Changes in the level of a time series are usually attributed to an intervention that affects its temporal evolution. The resulting time series are referred to as interrupted time series and may be used to identify the events that caused the intervention and to quantify their impact. In the present paper a heuristic method for level change detection in time series is presented. The method uses higher order statistics, namely the skewness and the kurtosis, and can identify both the existence of a change in the level of the time series as well as the instance it has happened. The technique is straightforward applicable to the detection of outliers in time series and promises to have several applications. The method is tested with both simulated and real world data and is compared to other popular change detection techniques.


 

A Recurrent Neural Network-based Forecasting System for Telecommunications Call Volume

 

P. Mastorocostas, C. Hilas, D. Varsamis, S. Dova

Applied Mathematics & Information Sciences, vol. 7, no 5, pp. 1643-1650, September 2013

 

Abstract

A recurrent neural-network based forecasting system for tecommunications call volume is proposed in this work. In particular, the forecaster is a Block-Diagonal Recurrent Neural Network with internal feedback. Model’s performance is evaluated by use of real-world telecommunications data, where an extensive comparative analysis with a series of existing forecasters is conducting, including both traditional models as well as neural and fuzzy approaches.


 

SCOLS-FuM: A Hybrid Fuzzy Modeling Method for Telecommunications Time-Series

 

P. Mastorocostas, C. Hilas

Informatica, vol. 25, no 2, pp. 221-239, 2014

 

Abstract

An application of fuzzy modeling to the problem of telecommunications time-series prediction is proposed in this paper. The model building process is a two-stage sequential algorithm, based on Subtractive Clustering (SC) and the Orthogonal Least Squares (OLS) techniques. Particularly, the SC is first employed to partition the input space and determine the number of fuzzy rules and the premise parameters. In the sequel, an orthogonal estimator determines the input terms which should be included in the consequent part of each fuzzy rule and calculate their parameters. A comparative analysis with well-established forecasting models is conducted on real world telecommunications data, where the characteristics of the proposed forecaster are highlighted.


 

ReNFFor: A Recurrent Neurofuzzy Forecaster for Telecommunications Data

 

P. Mastorocostas, C. Hilas

Neural Computing and Applications, vol. 22, iss. 7-8, pp. 1727-1734, June 2013.

 

Abstract

In this work a dynamic neurofuzzy system for forecasting outgoing telephone calls in a University Campus is proposed. The system comprises modified Takagi-Sugeno-Kang fuzzy rules, where the rules’ consequent parts are small neural networks with unit internal recurrence. The characteristics of the proposed forecaster, which is entitled Recurrent Neurofuzzy Forecaster (ReNFFor), are depicted via a comparative analysis with a series of well-known forecasting models.


 

A Computational Intelligence-based Forecasting System for Telecommunications Time Series

 

P. Mastorocostas, C. Hilas

Engineering Applications of Artificial Intelligence, vol. 25, iss. 1, pp. 200-206, February 2012

 

Abstract

In this work a computational intelligence-based approach is proposed for forecasting outgoing telephone calls in a University Campus. A modified Takagi-Sugeno-Kang fuzzy neural system is presented, where the consequent parts of the fuzzy rules are neural networks with internal recurrence, thus introducing dynamics to the overall system. The proposed model, entitled Locally Recurrent Neurofuzzy Forecasting System (LR-NFFS), is compared to well-established forecasting models, where its particular characteristics are highlighted.


 

A Block-Diagonal Recurrent Fuzzy Neural Network for System Identification

 

P. Mastorocostas, C. Hilas

Neural Computing and Applications, vol. 18, no 7, pp. 707-717, October 2009

 

Abstract

A recurrent fuzzy neural network with internal feedback is suggested in this paper. The network is entitled Dynamic Block-Diagonal Fuzzy Neural Network (DBD-FNN), and constitutes a generalized Takagi-Sugeno-Kang fuzzy system, where the consequent parts of the fuzzy rules are small Block-Diagonal Recurrent Neural Networks. The proposed model is applied to a benchmark identification problem, where a dynamic system is to be identified. Additionally, an application of the proposed model to the problem of the analysis of lung sounds is presented. Particularly, a filter based on the DBD-FNN is developed, trained with the RENNCOM method. Extensive experimental and simulation results are given and performance comparisons with a series of other models are conducted, highlighting the modeling characteristics of DBD-FNN as an identification tool and the effectiveness of the proposed separation filter.


 

An Application of Supervised and Unsupervised Learning Approaches to Telecommunications Fraud Detection

 

C. Hilas, P. Mastorocostas

Knowledge-Based Systems, vol. 21, iss. 7, pp. 721-726, October 2008

 

Abstract

This paper investigates the usefulness of applying different learning approaches to a problem of telecommunications fraud detection. Five different user models are compared by means of both supervised and unsupervised learning techniques, namely the multilayer perceptron and the hierarchical agglomerative clustering. One aim of the study is to identify the user model that best identifies fraud cases. The second task is to explore different views of the same problem and see what can be learned form the application of each different technique. All data come from real defrauded user accounts in a telecommunications network. The models are compared in terms of their performances. Each technique’s outcome is evaluated with appropriate measures.


 

A Pipelined Recurrent Fuzzy Model for Real-Time Analysis Lung Sounds

 

P. Mastorocostas, D. Stavrakoudis, J. Theocharis

Engineering Applications of Artificial Intelligence, vol.21, iss. 8, pp. 1301-1308, December 2008

 

Abstract

This paper presents a recurrent fuzzy-neural filter that performs the task of separation of lung sounds, obtained from patients with pulmonary pathology. The filter is a pipelined Takagi-Sugeno-Kang recurrent fuzzy network, consisting of a number of modules interconnected in a cascaded form. The participating modules are implemented through recurrent fuzzy neural networks with internal dynamics. The structure of the modules is evolved sequentially from input-output data. Extensive experimental results, regarding the lung sound category of crackles, are given, and a performance comparison with a series of other fuzzy and neural filters is conducted, underlining the separation capabilities of the proposed filter.


 

Simulated Annealing Dynamic RPROP for Training Recurrent Fuzzy Systems

 

P. Mastorocostas

Advances in Fuzzy Sets and Systems, vol. 2, iss. 3, pp. 283-300, October 2007

 

Abstract

An adaptive learning method, the SA-DRPROP (Simulated Annealing Dynamic Resilient Back Propagation), is proposed in this paper, for training recurrent fuzzy systems. The method modifies the SARPROP algorithm, originally developed for static neural models, in order to be applied to dynamic models. The SA-DRPROP qualities are investigated by a series of simulation examples, including a noise cancellation problem. Comparisons with other learning algorithms are given and discussed, indicating the enhanced learning capabilities of the proposed algorithm.


 

A Generalized Takagi-Sugeno-Kang Recurrent Fuzzy-Neural Model for Adaptive Noise Cancellation

 

P. Mastorocostas, D. Varsamis, C. Hilas, C. Mastorocostas

Neural Computing and Applications, vol. 17, no 5-6, pp. 521-529, October 2008

 

Abstract

This paper presents a recurrent fuzzy-neural filter for adaptive noise cancellation. The cancellation task is transformed to a system-identification problem, which is tackled by use of the Dynamic Neuron-based Fuzzy Neural Network. The fuzzy model is based on Takagi-Sugeno-Kang fuzzy rules, whose consequent parts consist of linear combinations of dynamic neurons. The Orthogonal Least Squares method is employed to select the number of rules, along with the number and kind of dynamic neurons that participate in each rule. Extensive simulation results are given and performance comparison with a series of other dynamic fuzzy and neural models is conducted, underlining the effectiveness of the proposed filter and its superior performance over its competing rivals.


 

A Locally Recurrent Globally Feed-forward Fuzzy Neural Network for Processing Lung Sounds

 

P. Mastorocostas, D. Varsamis, C. Mastorocostas, C. Hilas

Lecture Notes in Computer Science, vol. 4669, pp. 120-128, 2007

 

Abstract

This paper presents a locally recurrent globally feedforward fuzzy neural network, with internal feedback, that performs the task of separation of lung sounds, obtained from patients with pulmonary pathology. The filter is applied to the problem of separating in real-time adventitious discontinuous sounds, belonging to the lung sound category of squawks, from vesicular sounds. The characteristics of the proposed filter are highlighted via a performance comparison with a series of other recurrent and static fuzzy and neural filters.


 

A Dynamic Fuzzy Model for Processing Lung Sounds

 

P. Mastorocostas, D. Varsamis, C. Mastorocostas, C. Hilas

IEE Electronics Letters, vol. 43, iss. 6, pp. 320-322, March 2007

 

Abstract

This paper presents a dynamic fuzzy filter, with internal feedback, that performs the task of separation of lung sounds, obtained from patients with pulmonary pathology. The filter is a novel generalized TSK fuzzy model, where the consequent parts of the fuzzy rules are Block-Diagonal Recurrent Neural Networks. Extensive experimental results, regarding the lung sound category of coarse crackles, are given, and a performance comparison with a series of other fuzzy and neural filters is conducted, underlining the separation capabilities of the proposed filter.


 

A Dynamic Fuzzy-Neural Filter for Separation of Discontinuous Adventitious Sounds from Vesicular Sounds

 

P. Mastorocostas, J. Theocharis

Computers in Biology and Medicine, vol. 57, pp. 60-69, 2007

 

Abstract

This paper presents a recurrent filter that performs real-time separation of discontinuous adventitious sounds from vesicular sounds. The filter uses two Dynamic Fuzzy Neural Networks, operating in parallel, to perform the task of separation of the lung sounds, obtained from patients with pulmonary pathology. Extensive experimental results, including fine/coarse crackles and squawks, are given, and a performance comparison with a series of other models is conducted, underlining the separation capabilities of the proposed filter and its improved performance with respect to its competing rivals.


 

A Recurrent Fuzzy Filter for the Analysis of Lung Sounds

 

P. Mastorocostas

Fuzzy Sets and Systems, vol. 157, iss. 4, pp. 578-594, February 2006

 

Abstract

This paper presents a recurrent fuzzy-neural filter for real-time separation of discontinuous adventitious sounds from vesicular sounds. The filter uses two Dynamic Neuron-based Fuzzy Neural Networks to perform the task of separation. The networks are generated by the Dynamic Orthogonal Least Squares Method and are applied to all kinds of lung sounds. Extensive experimental results are given and a performance comparison with a series of other models is conducted, underlining the effectiveness of the proposed filter and its superior performance over its competing rivals.


 

A Stable Learning Algorithm for Block-Diagonal Recurrent Neural Networks: Application to the Analysis of Lung Sounds

 

P. Mastorocostas, J. Theocharis

IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics, vol. 36, no 2, pp.242-254, April 2006

 

Abstract

A novel learning algorithm, the RENNCOM (Recurrent Neural Network Constrained Optimization Method) is suggested in this paper, for training block-diagonal recurrent neural networks. The training task is formulated as a constrained optimization problem, whose objective is twofold: (i) minimization of an error measure, leading to successful approximation of the input/output mapping and (ii) optimization of an additional functional, the pay-off function, which aims at ensuring network stability throughout the learning process. Having assured the network and training stability conditions, the pay-off function is switched to an alternative form with the scope to accelerate learning. Simulation results on a benchmark identification problem demonstrate that, compared to other learning schemes with stabilizing attributes, the RENNCOM algorithm has enhanced qualities, including, improved speed of convergence, accuracy and robustness. The proposed algorithm is also applied to the problem of the analysis of lung sounds. Particularly, a filter based on block-diagonal recurrent neural networks is developed, trained with the RENNCOM method. Extensive experimental results are given and performance comparisons with a series of other models are conducted, underlining the effectiveness of the proposed filter.


 

Resilient Back Propagation Learning algorithm for Recurrent Fuzzy Neural Networks

 

P. Mastorocostas

IEE Electronics Letters, vol. 40, no 1, pp. 57-58, January 2004

 

Abstract

An efficient training method for recurrent fuzzy neural networks is proposed. The method modifies the RPROP algorithm, originally developed for static neural networks, in order to be applied to dynamic systems. A comparative analysis with the standard Back Propagation Through Time is given, indicating the effectiveness of the proposed algorithm.


 

An Orthogonal Least Squares Method for Recurrent Fuzzy-Neural Modeling

 

P. Mastorocostas, J. Theocharis

Fuzzy Sets and Systems, vol. 140, iss. 2, pp. 285-300, December 2003

 

Abstract

This paper presents an Orthogonal Least Squares (OLS) based modeling method, named Dynamic OLS (D-OLS), for generating recurrent fuzzy models. A Dynamic Neuron-based Fuzzy Neural Network (DN-FNN) is proposed, comprising generalized Takagi-Sugeno-Kang fuzzy rules, whose consequent parts consist of dynamic neurons with local output feedback. From an arbitrarily large set of candidate dynamic neurons, the D-OLS method selects automatically the most important ones. Thus, each fuzzy rule of the resulting model contains a different number and kind of dynamic neurons. The proposed dynamic model, equipped with the learning algorithm, is applied to two temporal problems, where the effectiveness of the suggested method as well as the advantages of the resulting dynamic model are demonstrated.


 

A Recurrent Fuzzy-Neural Model for Dynamic System Identification

 

P. Mastorocostas, J. Theocharis

IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics, vol. 32, no 2, pp. 176-190, April 2002

 

Abstract

This paper presents a fuzzy modeling approach for identification of dynamic systems. In particular, a new fuzzy model, the Dynamic Fuzzy Neural Network (DFNN), consisting of recurrent TSK rules, is developed. The premise and defuzzification parts are static while the consequent parts of the fuzzy rules are recurrent neural networks with internal feedback and time delay synapses. The network is trained by means of a novel learning algorithm, named Dynamic-Fuzzy Neural Constrained Optimization Method (D-FUNCOM), based on the concept of constrained optimization. The proposed algorithm is general since it can be applied to locally as well as fully recurrent networks, regardless of their structures. An adaptation mechanism of the maximum parameter change is presented as well. The proposed dynamic model, equipped with the learning algorithm, is applied to several temporal problems, including modeling of a NARMA process and the noise cancellation problem. Performance comparisons are conducted with a series of static and dynamic systems and some existing recurrent fuzzy models. Simulation results show that DFNN compares favorably with its competing rivals and thus it can be considered for efficient system identification.


 

A Constrained Orthogonal Least Squares Method for Generating TSK Fuzzy Models:

Application to Short-Term Load Forecasting

 

P. Mastorocostas, J. Theocharis, V. Petridis

Fuzzy Sets and Systems, vol. 118, iss. 2, pp. 35-53, March 2001

 

Abstract

In this paper, an Orthogonal Least Squares (OLS) based modeling method is developed, named the Constrained OLS (C-OLS), for generating simple and efficient TSK fuzzy models. The method is a two stage model building technique, where both premise and consequent identification are simultaneously performed. The fuzzy system is considered as a linear regression model by decomposing the TSK model into a collection of generic rules. The C-OLS algorithm is employed at stage-1 to identify the structure of the model. Given a model building data set, the algorithm selects a subset of most significant regressors which should be included in the model. Based on the similarity measure, a classification tool is developed, which organizes the selected terms into groups with similar premise parts, forming TSK rules. Additionally, input variable selection for the consequent part is performed. The resulting model is reduced in complexity by discarding the unnecessary terms, and is optimized at stage-2 using a richer training data set. This method is used to generate fuzzy models for a real-world problem, the load forecasting of the Greek power system. Extensive simulation results are given and discussed, demonstrating the effectiveness of the suggested method.


 

An Orthogonal Least Squares-Based Fuzzy Filter for Real-Time Analysis of Lung Sounds

 

P.A Mastorocostas, J.B Theocharis, Y.A. Tolias, L.J. Hadjileontiadis, S.M. Panas

IEEE Transactions on Biomedical Engineering, vol. 47, no 9, pp. 1165-1176, September 2000

 

Abstract

Pathological discontinuous adventitious sounds (DAS) are strongly related with the pulmonary dysfunction. Its clinical use for the interpretation of respiratory malfunction depends on their efficient and objective separation from vesicular sounds (VS). In this paper, an automated approach to the isolation of DAS from VS, based on their nonstationarity, is presented. The proposed scheme uses two fuzzy inference systems (FIS's), operating in parallel, to perform more flexibly the task of adaptive separation, resulting in the Orthogonal Least Squares-based Fuzzy Filter (OLS-FF). By applying the OLS-FF to fine/coarse crackles and squawks, selected from three lung sound databases, the coherent structure of DAS is revealed and they are efficiently separated from VS. When compared with previous works, the OLS-FF performs quite similarly, but with significantly lower computational load, resulting in a faster real-time clinical screening of DAS.


 

FUNCOM: A Constrained Learning Algorithm for Fuzzy Neural Networks

 

P. Mastorocostas, J. Theocharis

Fuzzy Sets and Systems, vol. 112, iss. 1, pp. 1-26, May 2000

 

Abstract

A novel learning algorithm, the FUNCOM (Fuzzy Neural Constrained Optimization Method) is suggested in this paper, for training fuzzy neural networks. The training task is formulated as a constrained optimization problem, whose objective is twofold: (i) minimization of an error measure, leading to successful approximation of the input/output mapping and (ii) optimization of an additional functional, which aims at formulating suitable internal representations of the fuzzy model. Optimization of the above functionals is carried out under the constraints imposed by the fuzzy system, which appear in the form of state equations. A fuzzy adaptation scheme is also suggested, which continuously modifies the step size during training, with the scope to improve the learning attributes of the algorithm. The FUNCOM qualities are investigated by a series of simulation examples. Comparisons with other learning algorithms are given and discussed, indicating the effectiveness of the proposed algorithm.


 

A Hybrid Fuzzy Modeling Method for Short Term Load Forecasting

 

P.A. Mastorocostas, J.B. Theocharis, S.J. Kiartzis, A.G. Bakirtzis

Mathematics and Computers in Simulation, vol. 51, iss. 3-4, pp. 221-232, January 2000

 

Abstract

A hybrid fuzzy modeling method is suggested in this paper, for short-term load forecasting. The method copes with both structure and parameter identification issues: first, the OLS method has been employed to perform the premise partitioning, thus to provide the initial premise parts of the fuzzy rules. In the sequel, the training task has been formulated as a two-stage optimization problem; at the first stage the consequent parameters are determined using the RLSE method while at the second stage a steepest ascent algorithm is employed to tune the premise parameters. The suggested method has been applied to the problem of short term load forecasting for the Greek power system, where different fuzzy models have been assigned to each day of the week for every season. Test results show that the developed STLF models produce accurate load predictions.


 

Fuzzy Modeling for Short Term Load Forecasting Using the Orthogonal Least Squares Method

 

P.A. Mastorocostas, J.B. Theocharis, A.G. Bakirtzis

IEEE Transactions on Power Systems, vol. 14, no 1, pp. 29-36, February 1999

 

Abstract

A fuzzy modeling method is developed in this paper for short term load forecasting. According to this method, identification of the premise part and consequent part is separately accomplished via the Orthogonal Least Squares (OLS) technique. Particularly, the OLS is first employed to partition the input space and determine the number of fuzzy rules and the premise parameters. In the sequel, a second orthogonal estimator determines the input terms, which should be included in the consequent part of each fuzzy rule and calculate its parameters. Input selection is automatically performed, given an input candidate set of arbitrary size, formulated by an expert. A satisfactory prediction performance is attained as shown in the test results, showing the effectiveness of the suggested method.


 

On the Nonlinear Behavior of the Analog Phase-Locked Loop: Synchronization

 

N. Margaris, P. Mastorocostas

IEEE Transactions on Industrial Electronics, vol. 43, no. 6, pp. 621-629, December 1996

 

Abstract

The synchronization, in the presence of time delay, of the nonlinear analog phase-locked loop (PLL) with an analog multiplier as phase detector (PD) and a lag filter is investigated. A nonlinear model for the voltage-controlled oscillator (VCO) is suggested and the sum frequency component at the PD output is taken into account. Simple expressions of the hold-in range of both the main synchronization and the synchronization at the third harmonic are derived. These expressions point out the effect of the time delay and the filter time constant on the hold-in range. Some conclusions of the presented analysis are not anticipated by the PLL classic theory and allow a better understanding of the loop behavior.


 

Conference Papers


 

A Graph Neural Network Based Learning Model for Urban Metro Flow Prediction

 

I. Drosouli, A. Voulodimos, P. Mastorocostas, G. Miaoulis, D. Ghazanfarpour

Twenty second IEEE International Conference on Machine Learning and Applications, Florida, USA, December 2023

 

Abstract

Transport data with dynamic spatial-temporal dependencies elevates transportation flow forecasting to a significant issue for operational planning, managing passenger flow, and arranging for individual travel in a smart city. The task is challenging due to the composite spatial dependency on transportation networks and the non-linear temporal dynamics with mobility conditions changing over time. To address these challenges, we propose a Spatial-Temporal Graph Convolutional Recurrent Network that learns from both the spatial stations network data and time-series of historical mobility changes so as to predict urban metro flow at a future time. The model is based on Graph Convolutional Networks (GCN) and Long Short-Term Memory (LSTM) in order to further improve the estimation accuracy. Extensive experiments on a real-world dataset of Hangzhou metro system prove the effectiveness of the proposed model.


 

Deep Learning Electric Load Forecasting for the Greek Power System

 

V. Zelios, P. Mastorocostas, P. Kandilogiannakis, A. Kesidis, P. Tselenti, A. Voulodimos

Twenty seventh International Conference on Circuits, Systems, Communications and Control, Rhodes, Greece, July 2023

 

Abstract

Long Short–Term Memory and Gated Recurrent Unit–based networks are examined for short–term load forecasting for the Greek power system, where predictions of the next hour’s load are made for the whole year. The recurrent property of the forecasters has the advantage to allow the use of a reduced input set since no prior knowledge of the most suitable previous loads is required for the model to identify the temporal relations that exist in the load time–series. An extensive simulation analysis is conducted, to evaluate the forecasters’ performance and to highlight their particular learning characteristics. Both deep learning–based forecasters perform very efficiently, with LSTM achieving an average percentage error of 1.18% and a root mean squared error of 83.8 MW, while GRU performs similarly with respective values of  1.17% and 83.4 MW.


 

A Dynamic Fuzzy Neural System for Time-Series Classification

 

P. Kandilogiannakis, P. Mastorocostas, C. Hilas

IEEE International Conference on Intelligent Systems, Varna, Bulgaria, August 2020

 

Abstract

A dynamic fuzzy neural system is proposed, for time series anomaly detection. The model is entitled BFuzzTiD (Block-diagonal Fuzzy Time-series Detector) and consists of fuzzy rules whose consequent parts are three-layer small recurrent neural networks. The hidden layer of each network has blocks of neurons that feed back to each other. BFuzzTid is trained by the Dynamic Resilient Propagation algorithm. The model learns the dynamics of the time series such that it can classify them by detecting the anomaly points. A comparative analysis is conducted with a series of time series anomaly detection models, in order to investigate the capabilities of the proposed detector.


 

ReFuzzTiD: A Recurrent Neurofuzzy Model for Anomaly Detection in Time Series

 

P. Kandilogiannakis, P. Mastorocostas

IEEE International Joint Conference on Neural Networks, Glasgow, UK, July 2020

 

Abstract

In this paper a recurrent neurofuzzy model is proposed, for time series anomaly detection. ReFuzzTiD comprises fuzzy rules whose consequent parts are simple three-layer neural networks with internal feedback at the neurons of the hidden layer. ReFuzzTid is trained by the Simulated Annealing Dynamic Resilient Propagation algorithm. The model learns the dynamics of the time series such that it can classify them by detecting the anomaly points. A comparative analysis with a series of time series anomaly detection models is given, highlighting the characteristics of the proposed detector.


 

A Computational Intelligence Model for Processing Lung Sounds

 

P. Mastorocostas, C. Hilas, J. Ellinas

International Conference on Signal and Image Processing, Oxford, UK, April 2019

 

Abstract

In this paper a computational intelligence-based filter is proposed, for real-time separation of the discontinuous adventitious lung sounds from the vesicular sounds. The filter is based on a Dynamic Block-Diagonal Fuzzy Neural Network to perform the task of separation of the lung sounds, obtained from patients with pulmonary pathology. The Simulated Annealing Dynamic Resilient Propagation algorithm is employed for training the neurofuzzy system, and the resulting filter is applied to two major classes of lung sounds. Extensive experimental results are given, along with a comparative analysis with a series of other filters, in order to highlight the separation potential of the proposed filter.


 

A Parallel Implementation of K-Means in Matlab

 

D. Varsamis, C. Talagkozis, A. Tsimpiris, P. Mastorocostas

Nineteenth International Conference on Advances in Distributed and Parallel Computing, Istanbul, Turkey, October 2017

 

Abstract

The aim of this work is the parallel implementation of k-means in Matlab, in order to reduce the execution time. Specifically, a new function in Matlab for serial k-means algorithm is developed, which meets all the requirements for the convertion to a function in Matlab with parallel computations. Additionally, two different variants for the definition of initial values are presented. In the sequel, the parallel approach is presented. Finally, the performance tests for the computation times respect to the numbers of features and classes are illustrated.


 

Exploiting the Interpretability of Fuzzy Rule-Based Classifiers for Analyzing Hyperspectral Remotely Sensed Data

 

D. Stavrakoudis, S. Mylonas, C. Topaloglou, J. Theocharis, P. Mastorocostas

Nineteenth International Conference on Circuits, Systems, Communications and Computers

Zakynthos, July 2015

 

Abstract

This paper showcases the use of fuzzy rule-based classification systems (FRBCSs) for analyzing hyperspectral data in the context of remote sensing classification tasks. First, the wavelet packet decomposition (WPD) is applied to the original data, in order to obtain higher-order features that describe the spectral changes within specific ranges of the electromagnetic spectrum. Subsequently, an advanced genetic FRBCS (GFRBCS) of the literature is applied, namely, the Fast Iterative Rule-based Linguistic Classifier (FaIRLiC), which is able to produce high-performing fuzzy rule bases with very low structural complexity. Ultimately, the information provided by FaIRLiC is exploited in order to discover useful hidden relations between the input features, which enable the easy discrimination between specific classes. As such, the employed model offers new tools for specialized photointerpretation, as well as the possibility for devising customized indices in the future. The procedure is employed here considering a Hyperion satellite image over a forested area in northern Greece, primarily focusing on the discrimination between two pine species.


 

Reconfigurable Hyper-Structures for Intrinsic Digital Circuit Evolution

 

S. Kazarlis, J. Kalomoiros, V. Kalaitzis, D. Bogas, P. Mastorocostas, A. Balouktsis, V. Petridis

Eighth International Conference on advances in Circuits, Electronics and Micro-Electronics

Venice, Italy, August 2015

 

Abstract

A workbench for intrinsic evolution of digital circuits is presented, based on a Cartesian Genetic Programming algorithm running on a personal computer and a reconfigurable platform suitable for run-time reconfiguration. Two types of Cartesian cell structures are proposed, based on a cylindrical interconnection grid. In addition to a feed-forward network, the cylindrical grid can allow feedback loops as well. The proposed structures are combined with dedicated communication and control logic, producing automatically a fitness result for each circuit configuration. The proposed system is tested with known digital circuits and evaluated in terms of resource usage and configuration speed.


 

A Method for Simulating Digital Circuits for Evolutionary Optimization

 

S. Kazarlis, J. Kalomoiros, P. Mastorocostas, V. Petridis, A. Balouktsis, V. Kalaitzis, A. Valais

Tenth International Joint Conferences on Systems, Computing Sciences and Software Engineering

December 2014

 

Abstract

This work presents a method for simulating asynchronous digital circuits, of both combinational and sequential logic, at the gate level. The simulator is going to serve as a fitness function of an Evolutionary Algorithm that will be used for optimal synthesis of digital circuits. Therefore the simulator needs to be simple, fast and reliable. The circuit under evaluation will be given to the simulator in an encoded form resembling DNA. Both the circuit codification method and the simulator are analytically discussed. Results are presented for a number of combinatorial and sequential digital circuits that prove the efficiency of the simulation method.


 

Accurate Crop Classification Using Hierarchical Fuzzy Rule-based Systems

 

C. Topalogloy, S. Mylonas, D. Stavrakoudis, J. Theocharis, P. Mastorocostas

2014 SPIE International Conference on Remote Sensing

Amsterdam, The Netherlands, September 2014

 

Abstract

This paper investigates the effectiveness of an advanced classification system for accurate crop classification using very high resolution (VHR) satellite imagery. Specifically, a recently proposed genetic fuzzy rule-based classification system (GFRBCS) is employed, namely, the Hierarchical Rule-based Linguistic Classifier (HiRLiC). HiRLiC’s model comprises a small set of simple IF–THEN fuzzy rules, easily interpretable by humans. One of its most important attributes is that its learning algorithm requires minimum user interaction, since the most important learning parameters affecting the classification accuracy are determined by the learning algorithm automatically. HiRLiC is applied in a challenging crop classification task, using a SPOT5 satellite image over an intensively cultivated area in a lake-wetland ecosystem in northern Greece. A rich set of higher-order spectral and textural features is derived from the initial bands of the (pan-sharpened) image, resulting in an input space comprising 119 features. The experimental analysis proves that HiRLiC compares favorably to other interpretable classifiers of the literature, both in terms of structural complexity and classification accuracy. Its testing accuracy was very close to that obtained by complex state-of-the-art classification systems, such as the support vector machines (SVM) and random forest (RF) classifiers. Nevertheless, visual inspection of the derived classification maps shows that HiRLiC is characterized by higher generalization properties, providing more homogeneous classifications that the competitors. Moreover, the runtime requirements for producing the thematic map was orders of magnitude lower than the respective for the competitors.


 

Spectral-Spatial Classification of Remote Sensing Images Using A Region-based GeneSIS

 

S. Mylonas, D. Stavrakoudis, J. Theocharis, P. Mastorocostas

2014 IEEE International Conference on Fuzzy Systems

Beijing, China, July 2014, pp. 1976-1984

 

Abstract

This paper proposes a spectral-spatial classification scheme for the classification of remotely sensed images, based on a new version of the recently proposed Genetic Sequential Image Segmentation (GeneSIS). GeneSIS segments the image in an iterative manner, whereby at each iteration a single object is extracted via a genetic algorithm-based object extraction method. In the previous version of GeneSIS, the candidate objects to be extracted were evaluated through the fuzzy content of their included pixels. In the present proposal, a watershed-driven fine segmentation map is initially obtained which serves as the basis for the upcoming GeneSIS segmentation. Our objective is to enhance the flexibility of the algorithm in extracting more flexible object shapes and reduce the execution time of the segmentation, while at the same time preserving all the inherent attributes of the GeneSIS procedure. Accordingly, the previously proposed fitness components are redefined in order to accommodate with the new structural components. In this work, the set of fuzzy membership maps required by GeneSIS are obtained via an unsupervised fuzzy clustering. The final classification result is obtained by combining the results from the unsupervised segmentation and the pixel-wise SVM classifier via majority voting. The validity of the proposed method is demonstrated on the land cover classification of a high-resolution hyperspectral image.


 

A Watershed-based Spectral-Spatial Segmentation and Classification Scheme for Remote Sensing Images

 

S. Mylonas, D. Stavrakoudis, C. Topaloglou, J. Theocharis, P. Mastorocostas

Fifth GEOBIA

Thessaloniki, May 2014, pp. 335-338

 

Abstract

In this paper, a new spectral-spatial classification scheme for the classification of remotely sensed images is presented. The method combines the results of supervised pixel-based classification with spatial information from unsupervised image segmentation. The image segmentation is performed in two stages. A fine segmentation map is initially obtained via watershed transformation. This region-based map serves as the basis for the next stage, where adjacent regions are merged according to the included clustering information. The final classification map is obtained via majority voting on the extracted segments. The validity of the proposed scheme is tested on the land cover classification of a hyperspectral image over a cultivated area.


 

A Genetic Programming Approach to Telecommunications Fraud Detection and Classification

 

C. Hilas, S. Kazarlis, I. Rekanos, P. Mastorocostas

International Conference on Circuits, Systems, Signal Processing, Communications and Computers

Venice, Italy, March 2014, pp. 77-83

 

Abstract

Telecommunications fraud has drawn the researchers’ attention due to the huge economic burden on companies and to the interesting aspect of users’ behavior modeling.  In the present paper, an application of genetic programming to fraud detection is presented.  Genetic programming is used for case classification in order to distinguish between normal and fraudulent activities in a telecommunications network. Implications to appropriate user behavior modeling are, also, discussed. Real world cases of defrauded user accounts are modeled by means of selected usage features and comparisons with other approaches are made.


 

A Block-Diagonal Recurrent Neural Network for Telecommunications Call Volume Forecasting

 

P. Mastorocostas, C. Hilas, D. Varsamis, S. Dova

International Conference on Computer Science and Electronics Engineering

Dubai, U.A.E., November 2013, ISBN: 9788192710419, pp. 7-11

 

Abstract

An application of recurrent neural modelling is proposed in this work. In particular, a forecaster based on the Block-Diagonal Recurrent Neural Network with internal feedback is applied to the problem of telecommunications call volume prediction. Model’s performance is evaluated by use of real-world telecommunications data, where a comparative analysis with a series of existing forecasters is conducting, including both traditional models as well as neural and fuzzy approaches.


 

A Telecommunications Call Volume Forecasting System Based on a Recurrent Fuzzy Neural Network

 

P. Mastorocostas, C. Hilas, D. Varsamis, S. Dova

2013 IEEE International Joint Conference on Neural Networks

Dallas, TX, U.S.A., August 2013, doi: 10.1109/IJCNN.2013.6707102

 

Abstract

The problem of telecommunications call volume forecasting is addressed to in this work. In particular, a foreacasting system is proposed, that is based on a dynamic fuzzy-neural model, where the consequent parts of the fuzzy rules are small Block-Diagonal Recurrent Neural Networks with internal feedback. The forecasting characteristics are highlighted and the prediction performance is evaluated by use of real-world telecommunications data. An extensive comparative analysis with a series of existing forecasters is conducting, including both traditional models as well as fuzzy and neurofuzzy approaches.


 

A Parallel Searching Algorithm for the Insetting Procedure in MATLAB Parallel Toolbox

 

D. Varsamis, P. Mastorocostas, A. Papakonstantinou, N. Karampetakis

Federated Conference on Computer Science and Information Systems

Wroclaw, Poland, September 2012, pp. 587-593

 

Abstract

In this paper we present the implementation of a parallel searching algorithm, which is used for the insetting procedure in cartography. The calculation time of the above procedure is very long due to the fact that the datasets in cartography are maps with large and very large resolution. The purpose of this proposal is to reduce the calculation time in a multicore machine with shared memory. The proposed algorithm and the performance tests are developed in Matlab Parallel Toolbox.


 

A TSK-based Fuzzy System for Telecommunications Time-Series Forecasting

 

P. Mastorocostas, C. Hilas, S. Dova, D. Varsamis

Sixth IEEE International Conference on Intelligent Systems

Sofia, Bulgaria, September 2012, pp. 146-151

 

Abstract

A two-stage model-building process for generating a Takagi-Sugeno-Kang fuzzy forecasting system is proposed in this paper. Particularly, the Subtractive Clustering (SC) method is first employed to partition the input space and determine the number of fuzzy rules and the premise parameters. In the sequel, an Orthogonal Least Squares (OLS) estimator determines the input terms which should be included in the consequent part of each fuzzy rule and calculate their parameters. A comparative analysis with well-established forecasting models is conducted on real world tele-communications data, in order to investigate the forecasting capabilities of the proposed scheme.


 

Forecasting of Telecommunications Time-Series via an Orthogonal Least Squares-based Fuzzy Model

 

P. Mastorocostas, C. Hilas, S. Dova, D. Varsamis

Twenty First IEEE International Conference on Fuzzy Systems

Brisbane, Australia, June 2012

 

Abstract

An application of fuzzy modeling to the problem of telecommunications data prediction is proposed in this paper. The model building process is a two-stage sequential algorithm, based on the Orthogonal Least Squares (OLS) technique. Particularly, the OLS is first employed to partition the input space and determine the number of fuzzy rules and the premise parameters. In the sequel, a second orthogonal estimator determines the input terms which should be included in the consequent part of each fuzzy rule and calculate their parameters. Input selection is automatically performed, given a large input candidate set. Real world telecommunications data are used in order to highlight the characteristics of the proposed forecaster and to provide a comparative analysis with well-established forecasting models.


 

Telecommunications Data Forecasting based on A Dynamic Neuro-Fuzzy Network

 

P. Mastorocostas, C. Hilas

Eighth International Symposium on Neural Networks

Guilin, China, May-June 2011, pp. 529-537

 

Abstract

In this work a dynamic neuro-fuzzy network (DyNF-Net) is proposed that is applied on the outgoing telephone traffic of a large organization. Based on the static type-1 fzzy system, it modifies its structure by introducing small recurrent neural networks with internal feedback. Real world telecommunications data are used in order to compare the proposed model with a series of existing forecasting models. The comparison highlights the particular characteristics of the proposed neuro-fuzzy network.


 

A Computational Intelligence Approach for Forecasting Telecommunication Time Series

 

P. Mastorocostas, C. Hilas

2010 International Conference on Telecommunications and Networks

December 2010, pp.585-596

 

Abstract

In this work a computational intelligence-based approach is proposed for forecasting outgoing telephone calls in a University Campus. A modified Takagi-Sugeno-Kang fuzzy neural system is presented, where the consequent parts of the fuzzy rules are neural networks with internal recurrence, thus introducing dynamics to the overall system. The proposed model, entitled Locally Recurrent Neurofuzzy Forecasting System (LR-NFFS), is compared to well-established forecasting models, where its particular characteristics are highlighted.


 

Change Level Detection in Time Series Using Higher Order Statistics

 

C. Hilas, I. Rekanos, S. Goudos, P. Mastorocostas, J. Sahalos

Sixteenth International Conferences on Digital signal Processing

Santorini, Greece, July 2009

 

Abstract

This paper presents a dynamic fuzzy filter with internal feedback, for adaptive noise cancellation. The cancellation task is transformed to a system-identification problem, which is tackled by the use of the Dynamic Block-Diagonal Fuzzy Neural Network. In order to underline the effectiveness of the proposed fuzzy noise canceller, it is applied to a benchmark noise cancellation problem, and a comparative analysis with a series of other dynamic models is conducted.


 

A Block-Diagonal Dynamic Fuzzy Filter for Adaptive Noise Cancellation

 

P. Mastorocostas, C. Hilas

2007 International Joint Conferences on Computer, Information, and Systems Sciences, and Engineering

December 2007, pp. 387-392

 

Abstract

This paper presents a dynamic fuzzy filter with internal feedback, for adaptive noise cancellation. The cancellation task is transformed to a system-identification problem, which is tackled by the use of the Dynamic Block-Diagonal Fuzzy Neural Network. In order to underline the effectiveness of the proposed fuzzy noise canceller, it is applied to a benchmark noise cancellation problem, and a comparative analysis with a series of other dynamic models is conducted.


 

A Block-Diagonal Recurrent Fuzzy Neural Network for Dynamic System Identification

 

P. Mastorocostas

Sixteenth IEEE International Conference on Fuzzy Systems

London, UK, July 2007, pp.11-16

 

Abstract

A recurrent fuzzy neural network with internal feedback is suggested in this paper. The network is entitled Dynamic Block-Diagonal Fuzzy Neural Network (DBD-FNN), and constitutes a generalized Takagi-Sugeno-Kang fuzzy system, where the consequent parts of the fuzzy rules are small Block-Diagonal Recurrent Neural Networks. The proposed model is applied to a benchmark problem, where a temporal system is to be identified. A comparative analysis with a series of recurrent fuzzy and neural models is conducted, highlighting the modeling characteristics of DBD-FNN.


 

A Pipelined Recurrent Fuzzy Neural Network for the Separation of Lung Sounds

 

D. Stavrakoudis, P. Mastorocostas, J. Theocharis

Sixteenth IEEE International Conference on Fuzzy Systems

London, UK, July 2007, pp. 49-54

 

Abstract

This paper presents a recurrent fuzzy-neural filter that performs the task of separation of lung sounds, obtained from patients with pulmonary pathology. The filter is a pipelined Takagi-Sugeno-Kang recurrent fuzzy network, consisting of a number of modules interconnected in a cascaded form. The participating modules are implemented through recurrent fuzzy neural networks with internal dynamics. The structure of the modules is evolved sequentially from input-output data. Extensive experimental results, regarding the lung sound category of crackles, are given, and a performance comparison with a series of other fuzzy and neural filters is conducted, underlining the separation capabilities of the proposed filter.


 

A Recurrent Neural Filter for Adaptive Noise Cancellation

 

P. Mastorocostas, D. Varsamis, C. Mastorocostas, C. Hilas

Fifth IASTED International Conference on Artificial Intelligence and Applications

Innsbruck, Austria, February 2006, pp. 341-346

 

Abstract

This paper presents a dynamic neural filter for adaptive noise cancellation. The cancellation task is transformed to a system-identification problem, which is tackled by use of the Block-Diagonal Recurrent Neural Network. The filter is applied to a benchmark noise cancellation problem, where a comparative analysis with a series of other dynamic models is conducted, underlining the effectiveness of the proposed filter and its superior performance over its competing rivals.


 

A Simulated Annealing-Based Learning Algorithm for Block-Diagonal Recurrent Neural Networks

 

P. Mastorocostas, D. Varsamis, C. Mastorocostas

Fifth IASTED International Conference on Artificial Intelligence and Applications

Innsbruck, Austria, February 2006, pp. 244-249

 

Abstract

A fast and efficient training method for block-diagonal recurrent fuzzy neural networks is proposed. The method modifies the Simulated Annealing RPROP algorithm, originally developed for static models, in order to be applied to dynamic systems. A comparative analysis with a series of algorithms and recurrent models is given, indicating the effectiveness of the proposed learning approach.


 

A Recurrent Fuzzy Filter for Adaptive Noise Cancellation

 

P. Mastorocostas, D. Varsamis, C. Mastorocostas, C. Hilas

CIMCA 2005

Vienna, Austria, November 2005, pp. 408-413

 

Abstract

This paper presents a recurrent fuzzy-neural filter for adaptive noise cancellation. The cancellation task is transformed to a system-identification problem, which is tackled by use of the Dynamic Neuron-based Fuzzy Neural Network. Extensive simulation results are given and performance comparison with a series of other dynamic fuzzy and neural models is conducted, underlining the effectiveness of the proposed filter and its superior performance over its competing rivals.


 

An Accelerating Learning Algorithm for Block-Diagonal Recurrent Neural Networks

 

P. Mastorocostas, D. Varsamis, C. Mastorocostas, I. Rekanos

CIMCA 2005

Vienna, Austria, November 2005, pp. 403-408

 

Abstract

An efficient training method for block-diagonal recurrent fuzzy neural networks is proposed. The method modifies the RPROP algorithm, originally developed for static models, in order to be applied to dynamic systems. A comparative analysis with a series of algorithms and recurrent models is given, indicating the effectiveness of the proposed learning approach.


 

A Constrained Optimization Algorithm for Training Locally Recurrent Globally Feedforward Neural Networks

 

P. Mastorocostas

2005 IEEE International Joint Conference on Neural Networks,

Montreal, Canada, July-August 2005, pp. 717-722

 

Abstract

This paper presents a novel learning algorithm for training locally recurrent globally feedforward neural networks. The training task is formulated as a constrained optimization problem, whose objective is twofold: (i) minimization of an error measure, leading to successful approximation of the input/output mapping and (ii) optimization of an additional functional, which aims at accelerating the learning process. Simulation results on a benchmark identification problem demonstrate that, compared to other learning schemes, the proposed algorithm has enhanced qualities, including improved speed of convergence, accuracy and robustness.


 

A Recurrent Fuzzy-Neural Filter for Real-Time Separation of Lung Sounds

 

P. Mastorocostas, J. Theocharis

2005 IEEE International Joint Conference on Neural Networks,

Montreal, Canada, July-August 2005, pp. 3023-3028

 

Abstract

This paper presents a recurrent filter that performs real-time separation of discontinuous adventitious sounds from vesicular sounds. The filter uses two Dynamic Fuzzy Neural Networks, operating in parallel, to perform the task of separation of the lung sounds, obtained from patients with pulmonary pathology. Extensive experimental results, including fine/coarse crackles and squawks, are given, and a performance comparison with a series of other models is conducted, underlining the separation capabilities of the proposed filter and its improved performance with respect to its competing rivals.


 

Simulated Annealing Dynamic RPROP For Training Recurrent Fuzzy Models

 

P. Mastorocostas, I. Rekanos

Fourteenth IEEE International Conference on Fuzzy Systems,

Reno, U.S.A., May 2005, pp. 1086-1091

 

Abstract

An adaptive learning method for recurrent fuzzy systems is proposed. The method modifies the SARPROP algorithm, originally developed for static neural models, in order to be applied to dynamic models. A comparative analysis with Dynamic RPROP and Back Propagation Through Time is given, indicating the enhanced learning capabilities of the proposed algorithm.


 

A Dynamic Fuzzy-Neural Filter for the Analysis of Lung Sounds

 

P. Mastorocostas, C. Chilas

2004 IEEE International Conference on Systems, Man, and Cybernetics,

Hague, the Netherlands, October 2004, pp.2231-2236

 

Abstract

This paper presents a recurrent fuzzy-neural filter for the separation of discontinuous adventitious sounds from vesicular sounds. The filter uses two Dynamic Neuron–based Fuzzy Neural Networks to perform the task of separation. The networks are generated by the Dynamic Orthogonal Least Squares Method and are applied to all kinds of lung sounds. Extensive experimental results are given and performance comparison with a series of other models is conducted, underlining the effectiveness of the proposed filter.


 

On Stable Learning of Block-Diagonal Recurrent Neural Networks

Part I: The RENNCOM Algorithm

 

P. Mastorocostas, J. Theocharis

2004 IEEE International Joint Conference on Neural Networks,

Budapest, Hungary, July 2004, pp. 815-820

 

Abstract

A novel learning algorithm, the RENNCOM (Recurrent Neural Network Constrained Optimization Method), is suggested in this paper, for training block-diagonal recurrent neural networks. The training task is formulated as a constrained optimization problem, whose objective is twofold: (i) minimization of an error measure, leading to successful approximation of the input/output mapping and (ii) optimization of an additional functional, which aims at ensuring network stability throughout the learning process. The characteristics of the proposed algorithm are highlighted by a simulation example, where a nonlinear dynamic identification problem is presented.


 

On Stable Learning of Block-Diagonal Recurrent Neural Networks

Part II: Application to the Analysis of Lung Sounds

 

P. Mastorocostas, J. Theocharis

2004 IEEE International Joint Conference on Neural Networks,

Budapest, Hungary, July 2004, pp.821-826

 

Abstract

A recurrent neural filter for the separation of discontinuous adventitious sounds from vesicular sounds is presented. The filter uses two Block-Diagonal Recurrent Neural Networks to perform the task of separation and is trained by the RENNCOM training algorithm. Extensive experimental results are given and performance comparisons with a series of other models are conducted, underlining the effectiveness of the proposed filter.


 

D-OLS: An Orthogonal Least Squares Method for Dynamic Fuzzy Models

 

P.A. Mastorocostas, J.B. Theocharis

Tenth IEEE International Conference on Fuzzy Systems,

Melbourne, Australia, December 2001, pp. 119-122

 

Abstract

This paper presents an Orthogonal Least Squares based modeling method, named Dynamic OLS (D-OLS), for generating recurrent fuzzy models. A Dynamic-Neuron based Fuzzy Neural Network (DN-FNN) is proposed, comprising generalized TSK fuzzy rules, whose consequent parts consist of dynamic neurons with local output feedback. From an arbitrarily large set of candidate dynamic neurons, the D-OLS method selects automatically the most important ones. Thus each fuzzy rule of the resulting model contains a different number and kind of dynamic neurons. In the simulation results, the effectiveness of the suggested method as well as the advantages of the resulting dynamic model are demonstrated.


 

A Generalized TSK Dynamic Fuzzy Neural network: Application to Adaptive Noise Cancellation

 

P.A. Mastorocostas, J.B. Theocharis

Ninth IEEE International Conference on Fuzzy Systems,

San Antonio, U.S.A., May 2000, pp. 877-882

 

Abstract

This paper presents a dynamic fuzzy neural network, consisting of generalized TSK rules. The premise and defuzzification parts are static while the consequent parts are recurrent neural networks with internal feedback and time delay synapses. The network is trained by means of a novel learning algorithm, based on the concept of constrained optimization. The suggested algorithm is general since it can be applied to locally as well as fully recurrent networks, regardless of their structure. An adaptation mechanism of the maximum parameter change is also developed. The proposed dynamic model, equipped with the learning algorithm, is employed as a noise cancellation filter, where it is compared with the ANFIS fuzzy filter. Simulation results show that the suggested model compares favorably with its comparing rival and can be regarded as a reliable, general adaptive filter.


 

Orthogonal Least Squares Based Fuzzy Model for Short Term Load Forecasting

 

P.A. Mastorocostas, J.B. Theocharis, S.J. Kiartzis, A.G. Bakirtzis

IMAC International Symposium on Soft Computing

Athens, Greece, June 1998

 

Abstract

This paper presents the development of a hybrid fuzzy modeling method for short-term load forecasting. The new approach employs the Orthogonal Least Squares method to create the fuzzy model and a constrained optimization algorithm to perform the parameter learning. The proposed model is tested using data of the Greek power system, while load forecasts with satisfying accuracy are reported.


 

Orthogonal Least Squares Fuzzy Modeling of Nonlinear Dynamical Systems

 

P. Mastorocostas, J. Theocharis

Sixth IEEE International Conference on Fuzzy Systems,

Barcelona, Spain, July 1997, pp. 1147-1152

 

Abstract

This paper suggests a fuzzy modeling technique that is based on the Orthogonal Least Squares Method (OLS). In particular, the OLS is employed to perform the partition of the input space and determine the number of fuzzy rules and the premise parameters. In the sequel, a second orthogonal estimator determines the terms that will be included in the consequent part of the fuzzy rules. The tuning of the selected consequent parameters is carried out by applying the Recursive Least Squares. The well-known gas furnace problem, given by Box and Jenkins, is used to illustrate the proposed modeling approach. Comparisons with other methods are given, where it is shown that the suggested method compares favorably with these modeling techniques.


 

FUNCOM: An Efficient Fuzzy Neural Training Algorithm

 

P. Mastorocostas, J. Theocharis

Proceedings, Fifth IEEE International Conference on Fuzzy Systems,

New Orleans, U.S.A., September 1996, pp. 380-386

 

Abstract

A novel algorithm for performing the parameter learning of Fuzzy Neural Networks is suggested, based on the concept of Constrained Optimization. The algorithm is applied to off-line and on-line training problems and is compared to standard Back Propagation and Fahlman’s Quickprop. The experimental results clearly show the efficiency of the suggested algorithm and its superiority over the comparing rivals.


 

A Fast Learning Hybrid Algorithm for Training Fuzzy Neural Networks

 

P. Mastorocostas, J. Theocharis

IASTED International Conference on Modelling, Identification and Control,

Innsbruck, Austria, February 1996, pp. 129-132

 

Abstract

A novel hybrid algorithm for performing the parameter learning of Fuzzy Neural Networks is suggested, employing a Constrained Optimization method and the Least Squares Estimate for determining the premise and consequent parameters, respectively. The algorithm is applied to off-line and on-line training problems and is compared to Quickprop, a faster variation of Back Propagation. The experimental results clearly show the efficiency of the suggested algorithm and its superiority over the comparing rival.