An Approach Using Particle Swarm Optimization and Rational Kernel for Variable Length Data Sequence Optimization
This paper proposes a novel approach for unsupervised classification of variable length sequence data using a concept inspired from the Particle Swarm Optimization and rational kernel. The approach uses the distance estimated by the rational kernel as a s
- PDF / 453,940 Bytes
- 9 Pages / 439.37 x 666.142 pts Page_size
- 47 Downloads / 297 Views
2
College of Engineering, Trivandrum, India [email protected] Computer Centre, University of Kerala, Trivandrum, India [email protected]
Abstract. This paper proposes a novel approach for unsupervised classification of variable length sequence data using a concept inspired from the Particle Swarm Optimization and rational kernel. The approach uses the distance estimated by the rational kernel as a similarity measure used for clustering the particles. It does not require the normalization of the data sequences into fixed size vectors. Each data sequence has a corresponding particle which moves in the parameter space towards other particles with the similar fitness value. Velocity factor which is used in updating particle position is influenced by the rational distance below a specified threshold. Experimental results display the robustness of proposed algorithm. Misclassification error for clustering the particles into different classes is provided in the results section. Keywords: Particle swarm optimization Variable length sequences Rational kernel Classification
1 Introduction Data classification and prediction are two popular sub areas of machine learning. Current algorithms in the sub field concentrates on data sets represented by fixed length feature vector to represent each sample data element. Popular applications from text classification and speech recognition to bioinformatics to information security require the classification of variable-length feature vector representing each data observations. Classifications of such variable length sequences normally employ principle of normalization of the variable length sequences into a fixed length vector before classification. This method suffers from serious drawback of creating degradation and aliasing which affects the accuracy of classification process. Particle Swarm optimization (PSO) is a nature inspired technique used to optimize a given problem iteratively based on fitness function [1]. PSO originated from the natural observation that a group of birds with similar requisite always flocks together to arrive at their objective. Coordinated group movements are achieved by the individual members by following general rules to try fly near the neighbor. Also keep distance with the neighbor to avoid collision with them neighbor. © Springer International Publishing Switzerland 2016 Y. Tan et al. (Eds.): ICSI 2016, Part I, LNCS 9712, pp. 401–409, 2016. DOI: 10.1007/978-3-319-41000-5_40
402
S. Raveendran and S.S. Vinodchandra
PSO has been popular and many versions of the algorithms have been proposed so as to make it suitable for different kind of optimization problems like unimodal functions, multi modal and multi objective optimization problems [2]. Parameter fine tuning is an important factor of the algorithm to make it suitable for specific applications [3]. The main disadvantage faced by all these versions is premature converge of the algorithm resulting in sub optimal solutions due to diversity in successive iterations. Another factor which affects the
Data Loading...
