Optimal SVM parameters estimation using chaotic accelerated particle swarm optimization for genetic data classification

Microarray studies and gene expression analysis have received tremendous attention over the last few years and provide many promising paths toward the understanding of fundamental questions in biology and medicine. High throughput biological data need to be processed, analyzed, and interpreted to address problems in life sciences. Feature selection (FS) and clustering are among the methods used in the statistical analysis of these data types. Support Vector Machine (SVM) is very popular and powerful among learning methods. In general, SVM has a wide range of applications in pattern recognition and nonlinear fitting. Kernel functions have an important role in classification capability of SVM. In this study, at first, selection of appropriate and efficient features is performed using Fisher ranking method considering classification power. After that, balancing is performed between different classes based on the removal of samples of the majority class, which are far from the decision boundary. Optimization of SVM classifier parameters including penalty factor C and Radial Basis Function (RBF) parameter γ is performed utilizing chaotic accelerated particle swarm optimization (CAPSO). Simulation is applied on genetic data sets. The classification accuracy of the proposed method on the incorporated genetic data sets is more than 94%. Using CAPSO algorithm instead of PSO algorithm for optimizing SVM parameters resulted in improved classification accuracy on genetic data.