Contents of Volume 26 (2016)

1/2016 2/2016 3/2016 4/2016 5/2016


  • [1] Krömer P., Misak S., Stuchly J., Platos J. (CZ)

    First page   Full text     DOI:

    Abstract: Precise wind energy potential assessment is vital for wind energy generation and planning and development of new wind power plants. This work proposes and evaluates a novel two-stage method for location-specific wind energy potential assessment. It combines accurate statistical modelling of annual wind direction distribution in a given location with supervised machine learning of efficient estimators that can approximate energy efficiency coefficients from the parameters of optimized statistical wind direction models. The statistical models are optimized using differential evolution and energy efficiency is approximated by evolutionary fuzzy rules.

  • Z. Zhang, A.K. Sangaiah (China, India),
    Guest Editorial of Special Issue on
    Full text

  • [2] Zhang Z., Han L., Li C., Wang J. (China)

    First page   Full text     DOI:

    Abstract: Multimedia social networks (MSNs) provide great convenience to users, while privacy leaks issues are becoming prominent. The studies on relationship-based access control have been widely used in social networks. However, with the dynamic development of social networks and rapid growth of user information, the access control does not completely meet the current system's need. In this paper, an attribute-based access control model called ABAC{MSN is proposed for MSNs. This model comprehensively considers user attributes, environment attributes and resource attributes, not only including relationships among users. In this model, users can set multimedia usage control policies based on three categories of user-defined attributes. A formal theoretical model is established, which includes constraint rules, data ow rules, policy conflict resolution mechanism, and applied to, a multimedia social-network-platform prototype systems. The deployment and application denote that this method effectively and exibly addresses use-case scenarios of multi-attribute-based media access control, and improves the access security of social media platforms and resources.

  • [3] Zhao C., Sun S., Han L., Peng Q. (China)

    First page   Full text     DOI:

    Abstract: Recommender systems have been well studied and applied both in the academia and industry recently. However, traditional recommender systems assume that all the users and items are independent and identically distributed. This assumption ignores the correlation of explicit attributes of both users and items. Aiming at modeling recommender systems more realistically and interpretably, we propose a novel and efficient hybrid matrix factorization method which combines implicit and explicit attributes, and can be used to solve the problem of cold start and recommender interpretation. Based on the MovieLens datasets, the experimental analysis shows our method is promising and efficient.

  • [4] Feng N., Yao Y. (China)

    First page   Full text     DOI:

    Abstract: The fuzzy morphological associative memories (FMAM) have many attractive advantages, but their recall effects for hetero associative memories are poor. This shortcoming impedes the application of hetero-FMAM. Aiming at the problem, and inspired by the unified framework of morphological associative memories, a new method called no rounding reverse fuzzy morphological associative memories (NR2FMAM) is presented by the paper. The value of the new method lies in hetero associative memories. Analyses and experiments show that, it has significantly affected hetero associative morphological memories and with a certain noise robustness. In some cases, it can work more effectively with greater correct recall rate than FMAM. The paper analyzes the reason that NR2FMAM is sometimes better than FMAM, and thinks that no rounding neural computing is one of passable reasons. At the same time, the condition that the recall rate of NR2FMAM is greater than FMAM is given by the corresponding theorem in this paper. The NR2FMAM not only enriched the theory of the morphological associative mnemonic framework, but also helps contribute to the solution of the hetero associative mnemonic problem which is incomplete. At the same time, it can serve as a new identification technology in social networks.

  • [5] L. Li, K. Fan, Z. Zhang, Z. Xia (China)

    First page   Full text     DOI:

    Abstract: Community structure implies some features in various real-world networks, and these features can help us to analysis structural and functional properties in the complex system. It has been proved that the classic k-means algorithm can efficiently cluster nodes into communities. However, initial seeds decide the efficiency of the k-means, especially when detecting communities with different sizes. To solve this problem, we improve the classic community detection algorithm with Principal Component Analysis (PCA) mapping and local expansion k-means. Since PCA can preserve the distance information of every node pairs, the improved algorithm use PCA to map nodes in the complex network into lower dimension European space, and then detect initial seeds for k-means using the improved local expansion strategy. Based on the chosen initial seeds, the k-means algorithm can cluster nodes into communities. We apply the proposed algorithm in real-world and artificial networks, the results imply that the improved algorithm is efficient to detect communities and is robust to the initial seed of K-means.

  • [6] Huang C., Li J., Dias S.M. (China)

    First page   Full text     DOI:

    Abstract: One focus of data analysis in formal concept analysis is attribute-significance measure, and another is attribute reduction. From the perspective of information granules, we propose information entropy in formal contexts and conditional information entropy in formal decision contexts, and they are further used to measure attribute significance. Moreover, an approach is presented to measure the consistency of a formal decision context in preparation for calculating reducts. Finally, heuristic ideas are integrated with reduction technique to achieve the task of calculating reducts of an inconsistent data set.

  • [7] Kaur S., Singh S., Kaushal S., Sangaiah A.K. (China)

    First page   Full text     DOI:

    Abstract: Web 2.0 has led to the expansion and evolution of web-based communities that enable people to share information and communicate on shared platforms. The inclination of individuals towards other individuals of similar choices, decisions and preferences to get related in a social network prompts the development of groups or communities. The identification of community structure is one of the most challenging task that has received a lot of attention from the researchers. Network community structure detection can be expressed as an optimisation problem. The objective function selected captures the instinct of a community as a group of nodes in which intra-group connections are much denser than inter-group connections. However, this problem often cannot be well solved by traditional optimisation methods due to the inherent complexity of network structure. Therefore, evolutionary algorithms have been embraced to deal with community detection problem. Many objective functions have been proposed to capture the notion of quality of a network community. In this paper, we assessed the performance of four important objective functions namely Modularity, Modularity Density, Community Score and Community Fitness on real-world benchmark networks, using Genetic Algorithm (GA). The performance measure taken to assess the quality of partitions is NMI (Normalized mutual information). From the experimental results, we found that the communities' identified by these objectives have different characteristics and modularity density outperformed the other three objective functions by uncovering the true community structure of the networks. The experimental results provide a direction to researchers on choosing an objective function to measure the quality of community structure in various domains like social networks, biological networks, information and technological networks.


  • [1] Ivan Nagy, Evgenia Suzdaleva (CZ)

    First page   Full text     DOI:

    Abstract: The paper deals with a problem of modeling discrete variables depending on continuous variables. This problem is known as the logistic regression estimated by numerical methods. The paper approaches the problem via the recursive Bayesian estimation of mixture models with the purpose of exploring a possibility of constructing the continuous data dependent switching model that should be estimated on-line. Here the model of the discrete variable dependent on continuous data is represented as the model of the mixture pointer dependent on data from mixture components via their parameters, which switch according to the activity of the components. On-line estimation of the data dependent pointer model has a great potential for tasks of clustering and classification. The specific subproblems include (i) the model parameter estimation both of the pointer and of the components obtained during the learning phase, and (ii) prediction of the pointer value during the testing phase. These two phases can be joined together in the case of necessity. A real-data experimental comparison with theoretical counterparts shows a competitiveness of the approach in the discussed field.

  • [2] Sung Ho Jang, Hyeok Gyu Kwon (Korea)

    First page   Full text     DOI:

    Abstract: The inferior cerebellar peduncle (ICP) is an important role in motor control, such as coordination of movement control of balance, posture, and gait. In the current study, using diffusion tensor tractography (DTT), we attempted to investigate the connectivity of the ICP in normal subjects. Forty healthy subjects were recruited for this study. DTTs were acquired using a sensitivity-encoding head coil at 1.5 Tesla. A seed region of interest was drawn at the ICP using the FMRIB Software Library. Connectivity was defined as the incidence of connection between the ICP and target brain regions at the threshold of 5, 25, and 50 streamlines. The ICP showed 100% connectivity to the vestibular nucleus, reticular formation, pontine tegmentum, and posterior lobe of the cerebellum, irrespective of thresholds. In contrast, the ICP showed more than 70% connectivity with the target brain regions at the threshold of 5 streamlines that is to the thalamus (100 %), anterior lobe of the cerebellum (100 %), pedunculopontine nucleus (95.0 %), red nucleus (92.5 %), primary somatosensory cortex (86.3 %), and primary motor cortex (75.0 %). According to our findings, the ICP had high connectivity, mainly with the sensory-motor related areas. We believe that the methodology and results of this study would be useful in investigation of the neural network associated with the sensory-motor system and brain plasticity following brain injury and other diseases.

  • [3] B. Yang, M. Xiang, Y.P. Zhang, S.M. Pang, X. Li (China)

    First page   Full text     DOI:

    Abstract: Feature reduction is an important issue in pattern recognition. Lower feature dimensionality could reduce the complexity and enhance the generalization ability of classifiers. In this paper we propose a new supervised dimensionality reduction method based on Locally Linear Embedding and Distance Metric Learning. First, in order to increase the interclass separability, a linear discriminant transformation learnt from distance metric learning is used to map the original data points to a new space. Then Locally Linear Embedding is adopted to reduce the dimensionality of data points. This process extends the traditional unsupervised Locally Linear Embedding to supervised scenario in a clear and natural way. In addition, it can also be seen as a general framework for developing new supervised dimensionality reduction algorithms by utilizing corresponding unsupervised methods. Extensive classification experiments performed on some real-world and artificial datasets show that the proposed method can achieve comparable to or even better results over other state-of-the-art dimensionality reduction methods.

  • [4] Yusuf Erzin, Mehdi Nikoo, Mohammad Nikoo, Tulin Cetin (Iran)

    First page   Full text     DOI:

    Abstract: In this study, the performance of three different self organization feature map (SOFM) network models denoted as SOFM1, SOFM2, and SOFM3 having neighborhood shapes, namely, SquareKohonenful, LineKohonenful, and Diamond-Kohenenful, respectively, to predict the critical factor of safety (Fs) of a widely-used artificial slope subjected to earthquake forces was investigated and compared. For this purpose, the reported data sets by Erzin and Cetin (2012) [7], including the minimum (critical) Fs values of the artificial slope calculated by using the simplified Bishop method, were utilized in the development of the SOFM models. The results obtained from the SOFM models were compared with those obtained from the calculations. It is found that the SOFM1 model exhibits more reliable predictions than SOFM2 and SOFM3 models. Moreover, the performance indices such as the determination coecient, variance account for, mean absolute error, root mean square error, and the scaled percent error were computed to evaluate the prediction capacity of the SOFM models developed. The study demonstrates that the SOFM1 model is able to predict the Fs value of the artificial slope, quite efficiently, and is superior to the SOFM2 and SOFM3.

  • [5] Dong Shao, Shenghai Hu, Yuting Fei (China)

    First page   Full text     DOI:

    Abstract: This paper proposes a new quantum particle swarm optimization algorithm with local attracting (LAQPSO), which is based on quantum-inspired evolutionary algorithm (QEA) and particle swarm optimization algorithm (PSO). In the proposed LAQPSO, a novel quantum bit expression mechanism called quantum angle is employed to encode the solution onto particle, and a new local attractor is proposed to determine the rotation angle of quantum rotation gate automatically. During the process of seeking the global solution, the magnitude of rotation angle is adjusted by an important parameter called contraction coefficient, which can quantitatively determine the tradeoff between exploration ability and exploitation ability. The simulation results for different contraction coecients are helpful for selecting the key parameter. A set of benchmark functions are used to evaluate the performance of LAQPSO, QEA and QBPSO, and the results show that the proposed algorithm has a fast convergence rate and can effectively avoid premature convergence.

  • [6] Petr Berka (CZ)

    First page   Full text     DOI:

    Abstract: Credit risk assessment, credit scoring and loan applications approval are one of the typical tasks that can be performed using machine learning or data mining techniques. From this viewpoint, loan applications evaluation is a classification task, in which the final decision can be either a crisp yes/no decision about the loan or a numeric score expressing the financial standing of the applicant. The knowledge to be used is inferred from data about past decisions. These data usually consist off both socio-demographic and economic characteristics of the applicant (e.g., age, income, and deposit), the characteristics of the loan, and the loan approval decision. A number of machine learning algorithms can be used for this purpose. In this paper we show how this task can be performed using the LISp- Miner system, a tool that is under development at the University of Economics, Prague. LISp-Miner is primarily focused on mining for various types of association rules, but unlike "classical" association rules proposed by Agrawal, LISp-Miner in- troduces a greater variety of different types of relations between the left-hand and right-hand sides of a rule. Two other procedures that can be used for classification task are implemented in LISp-Miner as well. We describe the 4ft-Miner and KEX procedures and show how they can be used to analyze data related to loan applications. We also compare the results obtained using the presented algorithms with results from standard rule-learning methods.


  • [1] Petr Vysoký (CZ)
    Prof. George J. Klir - Obituary, 331-333,       
    Full text

  • [2] Spera E., Migliore M., Unsworth N., Tegolo D. (Italy)
    On the cellular mechanisms underlying working memory capacity in humans, 335-350

    First page   Full text     DOI: 10.14311/NNW.2016.26.019

    Abstract: The cellular processes underlying individual differences in the Woring Memory Capacity (WMC) of humans are essentially unknown. Psychological experiments suggest that subjects with lower working memory capacity (LWMC), with respect to subjects with higher capacity (HWMC), take more time to recall items from a list because they search through a larger set of items and are much more susceptible to interference during retrieval. However, a more precise link between psychological experiments and cellular properties is lacking and very difficult to investigate experimentally. In this paper, we investigate the possible underlying mechanisms at the single neuron level by using a computational model of hippocampal CA1 pyramidal neurons, which have been suggested to be deeply involved in the recognition of specific items. The model makes a few experimentally testable predictions on the cellular processes underlying the cumulative latency in delayed free recall experimentally observed in humans under different testing conditions. The results suggest, for the first time, a physiologically plausible explanation for individual performances, and establish a proof of principle for the hypothesis that HWMC individuals use a larger portion of the apical tree with a correlated higher level of synaptic background noise.

  • [3] Douda V., Jánešová M. (CZ)
    Predictive model and methodology for optical telecommunications infrastructure, 351-362

    First page   Full text     DOI: 10.14311/NNW.2016.26.020

    Abstract: In this article a predictive model and a novel methodology of processing the data measured in the physical model of an optical telecommunications infrastructure is presented. The task is motivated by practical use of the results of experiments in the environment of the telecommunications network. We present an original predictive model and methodology, reflecting the specifics of examined infrastructure. The probabilistic prediction of the occurrence of emergencies is calculated via cluster analysis techniques used in Bayesian approach in the n-dimensional data space. The predictive model is experimentally verified on real data. Results of experiments are interpreted for practical use in real environment of the telecommunications infrastructure.

  • [4] Susi G., Cristini A., Salerno M. (Italy)
    Path multimodality in a Feedforward SNN module, using LIF with Latency model, 363-376

    First page   Full text     DOI: 10.14311/NNW.2016.26.021

    Abstract: In this paper, the network transmission properties of a feedforward Spiking Neural Network (SNN) affected by synchronous stimuli are investigated with respect to the connection probability and the synaptic strengths. By means of an event-driven method, all simulations are conducted using the Leaky Integrate-and-Fire with Latency (LIFL) model. Typical cases are taken into consideration, in which a network section (module) is able to process the input information, introducing a particular behavior, that we have called path multimodality. Simulation results are discussed. Through this phenomenon, the output layer of the network can generate a number of temporally spaced groups of synchronous spikes. The multimodality effect could be applied for various purposes, for instance in coding or else transmission issues.

  • [5] Xia M., Shen M., Wang J., Weng L., Yan C. (China)
    Anti-spurious-state neural network using nonlinear outer product and dynamic synapses, 377-392

    First page   Full text     DOI: 10.14311/NNW.2016.26.022

    Abstract: Associative memory (AM) is a very important part of the theory of neural networks. Although the Hebbian learning rule is always used to model the associative memory, it easily leads to spurious state because of the linear outer product method. In this work, nonlinear function constitution and dynamic synapses, against a spurious state for associative memory neural network are proposed. The model of the dynamic connection weight and the updating scheme of the states of neurons are presented. Nonlinear function constitution improves the conventional Hebbian learning rule to be a nonlinear outer product method. The simulation results show that both nonlinear function constitution and dynamic synapses can effectively enlarge the attractive basin. Comparing to the existing memory models, associative memory of neural network with nonlinear function constitution can both enlarge the attractive basin and increase the storage capacity. Owing to dynamic synapses, the attractive basin of the stored patterns is further enlarged, at the same time the attractive basin of the spurious state is diminished. But the storage capacity is decreased by using the dynamic synapses.

  • [6] Mosavi M.R., Khishe M., Ghamgosar A. (Iran)
    Classification of sonar data set using neural network trained by Gray Wolf Optimization, 393-415

    First page   Full text     DOI: 10.14311/NNW.2016.26.023

    Abstract: Multi-Layer Perceptron Neural Networks (MLP NNs) are the commonly used NNs for target classification. They purposes not only in simulated environments, but also in actual situations. Training such NNs has significant importance in a way that many researchers have been attracted to this field recently. Conventional gradient descent and recursive method has long been used to train NNs. Improper classification accuracy, slow convergence speed and trapping in local minimums are disadvantages of the traditional methods. In order to overcome these issues, in recent years heuristic and meta-heuristic algorithms are widely used. This paper uses Gray Wolf Optimization (GWO) algorithm for training the NN. This algorithm is inspired by lifestyle and hunting method of GWs. GWO has a superior ability to solve the high-dimension problems, so we try to classify the Sonar dataset using this algorithm. To test the proposed method, this algorithm is compared to Particle Swarm Optimization (PSO) algorithm, Gravitational Search Algorithm (GSA) and the hybrid algorithm (i.e. PSOGSA) using three sets of data. Measured metrics are convergence speed, the possibility of trapping in local minimum and classification accuracy. The results show that the proposed algorithm in most cases provides better or comparable performance compared to the other mentioned algorithms.


  • [1] Yudong Zhang and Bin Yu (China)
    Guest editorial, 203-204,       
    Full text

  • [2] M. Yang, C. Chen, L. Wang, X. Yan, L. Zhou (China)
    Bus Arrival Time Prediction using Support Vector Machine with Genetic Algorithm, 205-217

    First page   Full text     DOI: 10.14311/NNW.2016.26.011

    Abstract: Accurate prediction of bus arrival time is of great significance to improve passenger satisfaction and bus attraction. This paper presents the prediction model of bus arrival time based on support vector machine with genetic algorithm (GA-SVM). The character of the time period, the length of road, the weather, the bus speed and the rate of road usage are adopted as input vectors in Support Vector Machine (SVM), and the genetic algorithm search algorithm is combined to find the best parameters. Finally, the data from Bus No.249 in Shenyang, china are used to check the model. The experimental results show that the forecasting model is superior to the traditional SVM model and the Artificial Neural Network (ANN) model in terms of the same data, and is of higher accuracy, which verified the feasibility of the model to predict the bus arrival time.

  • [3] F. Guan, Z. Peng, K. Wang, X. Song, J. Gao (China)
    Multi-Step Hybrid Prediction Model of Baltic Supermax Index Based on Support Vector Machine, 219-232

    First page   Full text     DOI: 10.14311/NNW.2016.26.012

    Abstract: Accurate prediction of the Baltic index makes great difference to the strategic decision and risk avoidance of the enterprise. For the multi-step Baltic Supermax Index prediction, direct prediction and iterative prediction has its own advantages. Therefore, in this paper, in combination with direct and iterative prediction, based on Support Vector Machine (SVM), a hybrid multistep prediction model is put forward. In hybrid model, the output from the iterative model is a rough prediction and it need also be adjusted based on the output from the direct model. And weekly BSI data from January 2011 to November 2014 are used to test the model. The results show that the hybrid multistep prediction model based on SVM has high accuracy, and is feasible in the BSI prediction.

  • [4] S. Ozdemir, M. Demirtas, S. Aydin (Turkey)
    Harmonic Estimation Based Support Vector Machine for Typical Power Systems, 233-252

    First page   Full text     DOI: 10.14311/NNW.2016.26.013

    Abstract: The power quality in electrical energy systems is very important and harmonic is the vital criterion. Traditionally Fast Fourier Transform (FFT) and Discrete Fourier Transform (DFT) have been used for the harmonic distortion analysis and in the literature harmonic estimations have been made using different methods. As an alternative method, this paper suggested using Support Vector Machine (SVM) for harmonic estimation. The real power energy distribution system has been examined and the estimation results have been compared with measured real data. The proposed solution approach was comparatively evaluated with the ANN and LR estimation methods. Comparison results show that THD estimation values that were obtained by the SVM method are close to the THD estimation values obtained from ANN (Artificial Neural Network) and LR (Linear regression) methods. The numerical results clearly showed that the SVM method is valid for THD estimation in the power system.

  • [5] C.S Kanimozhiselvi, A. Pratap (India)
    Possibilistic LVQ neural network - An application to childhood autism grading, 253-269

    First page   Full text     DOI: 10.14311/NNW.2016.26.014

    Abstract: The nature of clinical diagnosis for psychological disorders are quiet different and difficult than the diagnosis of a disease. Generally they are assessed by screening certain behavioral features shown by the human that makes the differential diagnosis as a challenging task with respect to accuracy. This diagnostic reasoning process again becomes error prone when there are improper, insufficient clinical data and lack of clinical expertise. Thus artificial intelligence based assistances in predicting and assessing psychological disorders have gained much interest. Artificial intelligence based techniques like neural network can simulate expertise for supporting decision making problems in any domain. Childhood autism is a neuro-psychiatric developmental disorder that impairs mainly three functional areas in a child: social, communication and behavior. This article demonstrates the application of a Possibilistic- Linear Vector Quantization (Po- LVQ) neural network for the preliminary screening and grading of childhood autistic disorder. The diagnostic system assesses the grades as: ‘Normal’, ‘Mild-Moderate’, ‘Moderate-Severe’, ‘Severe’. It is able to perform with an improved overall accuracy of 95% exactly agreeing to the diagnostic criteria. Results of other performance parameters are also good enough to support the existing works about the applicability of neural network in autism diagnosis. Hence this research proposes a Po-LVQ based assessment support system for the diagnostic confirmation in grading childhood autism, during uncertain diagnosis due to lack of expertise. This helps to reduce the frustration and lengthy delays experiences to parents before obtaining an accurate diagnostic result.

  • [6] B. Yu, Y.T. Wang, J.B. Yao, J.Y. Wang (China)
    A Comparison of the Performace of ANN and SVM for the Prediction of Traffic Accident Duration, 271-287

    First page   Full text     DOI: 10.14311/NNW.2016.26.015

    Abstract: The prediction of traffic accident duration is great significant for rapid disposal of traffic accidents, especially for fast rescue of traffic accidents and re- moving traffic safety hazards. In this paper, two methods, which are based on artificial neural network (ANN) and support vector machine (SVM), are adopted for the accident duration prediction. The proposed method is demonstrated by a case study using data on approximately 235 accidents that occurred on freeways located between Dalian and Shenyang, from 2012 to 2014. The mean absolute error (MAE), the root mean square error (RMSE) and the mean absolute percentage error (MAPE) are used to evaluate the performances of the two measures. The conclusions are as follows: Both ANN and SVM models had the ability to predict traffic accident duration within acceptable limits. The ANN model gets a better result for long duration incident cases. The comprehensive performance of the SVM model is better than the ANN model for the traffic accident duration prediction.

  • [7] Ch. Ferles, A. Stafylopatis (Greece)

    First page   Full text     DOI: 10.14311/NNW.2016.26.016

    Abstract: The present study devises two techniques for visualizing biological sequence data clusterings. The Sequence Data Density Display (SDDD) and Sequence Likelihood Projection (SLP) visualizations represent the input symbolical sequences in a lower-dimensional space in such a way that the clusters and relations of data elements are preserved as faithfully as possible. The resulting unified framework incorporates directly raw symbolical sequence data (without necessitating any preprocessing stage), and moreover, operates on a pure unsupervised basis under complete absence of prior information and domain knowledge.

  • [8] P. Zahradnik, E.B. Aynakulov, R. Klof, B. Šimák (CZ, Kazakhstan)
    A Simple Echo Attenuation in Signals, 305-315

    First page   Full text     DOI: 10.14311/NNW.2016.26.017

    Abstract: A simple yet powerful procedure for an echo attenuation in signals is introduced. The presented method involves no external reference signal. It is based on comb FIR filtering. To the advantages of the described method belong the simplicity and performance which are beneficial in real time implementations. For illustration, a simulation of the procedure is included. The efficiency of the presented method is demonstrated by a real time implementation on a digital signal processor.

  • [9] J. Brandts, M. Křížek, Z. Zhang (NL, CZ, China/US)
    Paradoxes in Numerical Calculations, 317-330

    First page   Full text     DOI: 10.14311/NNW.2016.26.018

    Abstract: When solving problems of mathematical physics using numerical methods we always encounter three basic types of errors: modeling error, discretization error, and round-off errors. In this survey, we present several pathological examples which may appear during numerical calculations. We will mostly concentrate on the infl uence of round-off errors.


  • [1] J. Hlavica, M. Prauzek, T. Peterek, P. Musilek (CZ)
    Assessment of Parkinson's disease progression using neural network and ANFIS models, 111-128

    First page   Full text     DOI: 10.14311/NNW.2016.26.006

    Abstract: Patients suffering from Parkinson's disease must periodically undergo a series of tests, usually performed at medical facilities, to diagnose the current state of the disease. Parkinson's disease progression assessment is an important set of procedures that supports the clinical diagnosis. A common part of the diagnostic train is analysis of speech signal to identify the disease-specific communication issues. This contribution describes two types of computational models that map speech signal measurements to clinical outputs. Speech signal samples were acquired through measurements from patients suffering from Parkinson's disease. In addition to direct mapping, the developed systems must be able of generalization so that correct clinical scale values can be predicted from future, previously unseen speech signals. Computational methods considered in this paper are artificial neural networks, particularly feedforward networks with several variants of backpropagation learning algorithm, and adaptive network-based fuzzy inference system (ANFIS). In order to speed up the learning process, some of the algorithms were parallelized. Resulting diagnostic system could be implemented in an embedded form to support individual assessment of Parkinson's disease progression from patients' homes.

  • [2] C. Kosun, G. Tayfur, H.M. Celik (Turkey)
    Soft computing and regression modelling approaches for link-capacity functions, 129-140

    First page   Full text     DOI: 10.14311/NNW.2016.26.007

    Abstract: Link-capacity functions are the relationships between the fundamental traffic variables like travel time and the flow rate. These relationships are important inputs to the capacity-restrained traffic assignment models. This study investigates the prediction of travel time as a function of several variables V/C (flow rate/capacity), retail activity, parking, number of bus stops and link type. For this purpose, the necessary data collected in Izmir, Turkey are employed by Artificial Neural Networks (ANNs) and Regression-based models of multiple linear regression (MLR) and multiple non-linear regression (MNLR). In ANNs modelling, 70% of the whole dataset is randomly selected for the training, whereas the rest is utilized in testing the model. Similarly, the same training dataset is employed in obtaining the optimal values of the coefficients of the regression-based models. Although all of the variables are used in the input vector of the models to predict the travel time, the most significant independent variables are found to be V/C and retail activity. By considering these two significant input variables, ANNs predicted the travel time with the correlation coefficient R = 0.87 while this value was almost 0.60 for the regression-based models.

  • [3] A New Artificial Intelligence Optimization Method for PCA Based Unsupervised Change Detection of Remote Sensing Image Data, 141-154

    First page   Full text     DOI: 10.14311/NNW.2016.26.008

    Abstract: In this study, a new artificial intelligence optimization algorithm, Differential Search (DS), was proposed for Principal Component Analysis (PCA) based unsupervised change detection method for optic and SAR image data. The model firstly computes an eigenvector space using previously created k×k blocks. The change detection map is generated by clustering the feature vector as two clusters which are changed and unchanged using Differential Search Algorithm. For clustering, a cost function is used based on minimization of Euclidean distance between cluster centers and pixels. Experimental results of optic and SAR images proved that proposed approach is effective for unsupervised change detection of remote sensing image data.

  • [4] A. Akbarimajd, M. Selseleh Jonban, M. Nooshyar, M. Davari (Iran)
    Neural Network based identification of Trichoderma species, 155-174

    First page   Full text     DOI: 10.14311/NNW.2016.26.009

    Abstract: The genus Trichoderma acts as an important antagonist against phytopathogenic fungi. This paper proposes a software-based identification tool for recognition of different species of Trichoderma. The method uses the morphological features for identification. Morphological-based species recognition is common method for identifying fungi, but regarding the similarity of morphological features among different species, their manual identification is difficult, time-consuming and may bring about faulty results. In this paper it is intended to identify different species of Trichoderma by means of neural network. For this purpose, 14 characteristics are used including 5 macroscopic and 9 microscopic characteristics. After quantifying qualitative features and training a multilayer perceptron neural network with quantified data, 25 species of Trichoderma are recognized by using the network. Totally, identification of Trichoderma species as one useful fungus is achieved by using the trained network.

  • [5] S. Sasikala, S. Appavu, S. Geetha (India)
    Improving detection performance of artificial neural network by Shapley value embedded genetic feature selector , 175-201

    First page   Full text     DOI: 10.14311/NNW.2016.26.010

    Abstract: This work is motivated by the interest in feature selection that greatly affects the detection accuracy of a classifier. The goals of this paper are (i) identifying optimal feature subset using a novel wrapper based feature selection algorithm called Shapley Value Embedded Genetic Algorithm (SVEGA), (ii) showing the improvement in the detection accuracy of the Artificial Neural Network (ANN) classifier with the optimal features selected, (iii) evaluating the performance of proposed SVEGA-ANN model on the medical datasets. The medical diagnosis system has been built using a wrapper based feature selection algorithm that attempts to maximize the specificity and sensitivity (in turn the accuracy) as well as by employing an ANN for classification. Two memetic operators namely “include” and “remove” features (or genes) are introduced to realize the genetic algorithm (GA) solution. The use of GA for feature selection facilitates quick improvement in the solution through a fine tune search. An extensive experimental evaluation of the proposed SVEGA-ANN method on 26 benchmark datasets from UCI Machine Learning repository and Kent ridge repository, with three conventional classifiers, outperforms state-of-the-art systems in terms of classification accuracy, number of selected features and running time.


  • [1] Karel Segeth (CZ)
    Miroslav Fiedler passed away, 1-3,       
    Full text

  • [2] Zhang Y., Zheng C.-D. (China)
    Novel stochastic stability conditions of fuzzy neural networks with Markovian jumping parameters under impulsive perturbations, 7-28

    First page   Full text     DOI: 10.14311/NNW.2016.26.001

    Abstract: This paper investigates the stochastic stability of fuzzy neural networks with Markovian jumping parameters and mixed delays under impulsive per- turbations in mean square. The mixed delays consist of time-varying delay and continuously distributed delay. By employing a new Lyapunov-Krasovskii functional, linear convex combination technique, a novel reciprocal convex lemma and the free-weight matrix method, two novel sufficient conditions are derived to ensure the stochastic asymptotic stability of the equilibrium point of the considered networks in mean square. The proposed results, which are expressed in terms of linear matrix inequalities, can be easily checked via Matlab LMI Toolbox. Finally, two numerical examples are given to demonstrate the effectiveness and less conservativeness of our theoretical results over existing literature.

  • [3] García-Manso A., Gallardo-Caballero R., García-Orellana C.J., González-Velasco H.M., Macías-Macías M. (Spain)
    Diagnosing breast masses using ICA and non-image features, 29-44

    First page   Full text     DOI: 10.14311/NNW.2016.26.002

    Abstract: One of the most challenging task for Computer Aided Diagnosis (CADx) systems designed to diagnose breast cancer is to be able to differentiate between benign and malignant masses. In this work we present a study made as part of an ongoing project whose aim is to develop an image-based CADx system for diagnosing mass lesions. Our system is based on image-based and non-image features. Image-based features are obtained using Independent Component Analysis (ICA), and both age and mammogram density are tested as non-image features. Performance results are provided for all the valid masses in a public database, obtaining a statistically significant improvement by adding age to image-based features. However, the addition of the density of the mammogram does not improve the system performance.

  • [4] Barigou F. (Algeria)
    Improving K-nearest neighbor efficiency for text categorization, 45-66

    First page   Full text     DOI: 10.14311/NNW.2016.26.003

    Abstract: With the increasing use of the Internet and electronic documents, automatic text categorization becomes imperative. Many classification methods have been applied to text categorization. The k-nearest neighbors (k-NN) is known to be one of the best state of the art classifiers when used for text categorization. However, k-NN suffers from limitations such as high computation, low tolerance to noise, and its dependency to the parameter k and distance function. In this paper, we first survey some improvements algorithms proposed in the literature to face those shortcomings. And second, we discuss an approach to improve k-NN efficiency without degrading the performance of classification. Experimental results on the 20 Newsgroup and Reuters corpora show that the proposed approach increases the performance of k-NN and reduces the time classification.

  • [5] Adam A., Ibrahim Z., Mokhtar N., Shapiai M.I., Mubin M. (Malaysia)
    Evaluation of different peak models of eye blink EEG for signal peak detection using artificial neural network, 67-90

    First page   Full text     DOI: 10.14311/NNW.2016.26.004

    Abstract: There is a growing interest of research being conducted on detecting eye blink to assist physically impaired people for verbal communication and controlling devices using electroencephalogram (EEG) signal. One particular eye blink can be determined from use of peak points. Therefore, the purpose of peak detection algorithm is to distinguish an actual peak location from a list of peak candidates. The need of a good peak model is important in ensuring a satisfy classification performance. In general, there are various peak models available in literature, which have been tested in several peak detection algorithms. In this study, performance evaluation of the existing peak models is conducted based on Artificial Neural Network (ANN) with particle swarm optimization (PSO) as learning algorithm. This study evaluates the performance of eye blink EEG signal peak detection algorithm for four different peak models which are Dumpala's, Acir's, Liu's, and Dingle's peak models. To generalize the performance evaluation, two case studies of eye blink EEG signal are considered, which are single and double eye blink signals. It has been observed that the best test performance, in average, is 91.94% and 87.47% for single and double eye blink signals, respectively. These results indicate that the Acir's peak model offers high accuracy of peak detection for the two eye blink EEG signals as compared to other peak models. The result of statistical analysis also indicates that the Acir's peak model is better than Dingle's and Dumpala's peak models.

  • [6] Seenivasagam V., Chitra R. (India)
    Myocardial infarction detection using intelligent algorithms, 91-110

    First page   Full text     DOI: 10.14311/NNW.2016.26.005

    Abstract: Myocardial Infarction (MI) also known as heart attack is one of the most dangerous cardiovascular diseases. Accurate early prediction can effectively reduce the mortality rate caused by MI. The early stages of MI may only have subtle indications which can be varied in variable risk factors and making diagnosis difficult even for experienced cardiologists. In this paper the computer aided detection system is proposed to find the risk level of MI using the supervised classifier. The MI prediction system is developed using Feed Forward Neural Network (FFNN), Cascade Correlation Neural Network (CNN), and Support Vector Machine (SVM). Genetic Optimized Neural Network (GAANN), Particle Swarm Optimized Neural Network (PSONN) and also the performance of the Computer Aided Detection system is analyzed using various performance metrics.