Ilya Hodashinsky

Fuzzy classifier design using harmonic search methods

A new approach to design of a fuzzy-rule-based classifier that is capable of selecting informative features is discussed. Three basic stages of the classifier construction—feature selection, generation of fuzzy rule base, and optimization of the parameters of rule antecedents—are identified. At the first stage, several feature subsets on the basis of discrete harmonic search are generated by using the wrapper scheme. The classifier structure is formed by the rule base generation algorithm by using extreme feature values. The optimal parameters of the fuzzy classifier are extracted from the training data using continuous harmonic search. Akaike information criterion is deployed to identify the best performing classifiers. The performance of the classifier was tested on real-world KEEL and KDD Cup 1999 datasets. The proposed algorithms were compared with other fuzzy classifiers tested on the same datasets. Experimental results show efficiency of the proposed approach and demonstrate that highly accurate classifiers can be constructed by using relatively few features.

Takagi–Sugeno Fuzzy System to Predict Discharge Ignition Probability in the Foreline Pressure Range

For materials, science it is important to study the structure and behavior of matter at the deepest level. Currently, modern microscopes allow one to see the atomic structure of matter. Materials should be prepared in a special way, for research in the microscope, but thus the natural structure of the material may changed. Especially, the processes at the atomic level are difficult to explore. In a computer model of matter, one can account the properties of atoms and even its electron structure. In this paper, by molecular dynamics method the structure and evolution of the palladium and its compounds with hydrogen are investigated. In the work equilibrium structure of the crystal lattice of palladium was obtained and determined the equilibrium lattice parameter at different temperatures. The behavior of the hydrogen atom inside the crystal lattice of palladium was studied. The structural and diffusion properties system palladium-hydrogen were obtained

Identification of Fuzzy Classifiers Based on Weed Optimization Algorithm

A weed optimization algorithm for parameter identification of fuzzy classifiers is proposed. This algorithm is inspired from colonizing weeds, which is shown to be very robust and adaptive to changes in the environment. Weed optimization has been found to be a simple but powerful algorithm for function optimization over continuous spaces. It has reportedly outperformed many types of evolutionary algorithms and other search heuristics when tested over both benchmark and real-world problems. To initialize the classifiers structure, the k-means algorithm is used. The experimental results obtained on five datasets from the KEEL repository (iris, pima, thyroid, Wisconsin, and so- nar) are presented. The proposed algorithm is compared with the algorithms de- signed to solve the same problem. The comparison shows usability of the pro- posed approach for solving classification problems.

Takagi-Sugeno fuzzy systems structure identification based on piecewise linear initialization

A new method is proposed for structure identification of Takagi-Sugeno fuzzy systems, which is called piecewise linear initialization (PLI). This method is based on clustering of input data and has only one parameter: mean-square deviation of a hyperplane in the cluster from data. Each cluster is a particular rule of the fuzzy system. Based on the cluster are constructed Gaussian membership functions, otherwise consequents of fuzzy rules are constructed using the recursive least square method. The proposed method is compared with other methods by analyzing the mean-square error and the average number of rules on various datasets from the KEEL repository.

Feature Selection and Identification of Fuzzy Classifiers Based on the Cuckoo Search Algorithm

Classification is an important problem of data mining. The main advantage of fuzzy methods for extracting classification rules from empirical data is that the user can easily understand and interpret these rules, which makes fuzzy classifiers a useful modeling tool. A fuzzy classifier uses IF-THEN rules, with fuzzy antecedents (IF-part of the rule) and class labels in consequents (THEN-part of the rule). A method to constructing fuzzy classifiers based on the cuckoo search metaheuristic is described. The proposed method to constructing fuzzy classifiers based on observations data involves three stages: (1) feature selection, (2) structure generation, and (3) parameter optimization. The contributions of this paper are: (i) proposal of Cuckoo Search based feature selection; (ii) proposal of Cuckoo Search based parameter optimization of fuzzy classifier; (iii) proposal of subtractive clustering algorithm for structure generation of fuzzy classifier; and (iv) experiments with well-known benchmark classification problems (wine, vehicle, hepatitis, segment, ring, twonorm, thyroid, spambase reproduction data sets).

Tradeoff search methods between interpretability and accuracy of the identification fuzzy systems based on rules

This paper starts a brief historical overview of occurrence and development of fuzzy systems and their applications. Integration methods are proposed to construct a fuzzy system using other AI methods, achieving synergy effect. Accuracy and interpretability are selected as main properties of rule-based fuzzy systems. The tradeoff between interpretability and accuracy is considered to be the actual problem. The purpose of this paper is the in-depth study of the methods and tools to achieve a tradeoff for accuracy and interpretability in rule-based fuzzy systems and to describe our interpretability indexes. A comparison of the existing ways of interpretability estimation has been made We also propose the new way to construct heuristic interpretability indexes as a quantitative measure of interpretability. In the main part of this paper we describe previously used approaches, the current state and original authors’ methods for achieving tradeoff between accuracy and complexity.

Generalized inverse-based recurrent algorithm for TS fuzzy system identification

This paper presents a recurrent algorithm based on generalized inverses for identifying consequent parameters of Takagi-Sugeno (TS) fuzzy systems. For identification of antecedent parameters, the piecewise linear initialization (PLI) algorithm is used. The capabilities of the algorithms in solving data approximation problems are tested on 14 datasets from the KEEL repository.

Comparative analysis of feature selection algorithms in construction of fuzzy classifiers

The goal of the paper is the analysis of the fuzzy classifiers effectiveness, which are built by different algorithms of feature selection according to wrapper algorithms. The search of the informative features is provided on basis of the greedy algorithm (GrA), the discrete genetic algorithm (GA), the discrete mine blast algorithm (MBAd).

Fuzzy simulation system for finding optimal current densities in the electron beam of a forevacuum plasma source

This paper describes the two-stage construction of a fuzzy simulation system for finding optimal current densities in the electron beam of a forevacuum plasma source. The first stage consists in solving a direct problem of constructing a fuzzy model describing the dependence of the current density on the parameters of the plasma source. The second stage consists in solving an inverse problem of finding the parameters of the plasma source taking into account the variations in the current density along the cross section of the electron beam. The simulation is carried out based on the Takagi–Sugeno (TS) fuzzy system constructed using the piecewise linear initialization (PLI) method, cuckoo search (CS) algorithm, and recursive least squares (RLS) method.

Development of hybrid intelligent system of express-diagnostics for detection potential attacker

The paper describes development of an intelligent system based on number of various subfield of mathematics, information and computer science. Propose of the intelligent system is fast detection potential attacker of the information. That type of intelligent system called hybrid intelligent system. The basic of the intelligent system include the test pattern recognition, discrete mathematics, logic-combinatorial and logic-combinatorial-probabilistic algorithms, separating systems, the threshold and fuzzy logic, soft computing, specific fuzzy classifiers and cognitive tools. The paper describes the scheme of components interaction of hybrid intelligent system. The combination of those components provides effectiveness at detection potential attackers of the information better than these components separately.