A novel F-SVM based on FOA for improving SVM performance

Abstract

Parameter setting is critical for the solution efficiency and accuracy of support vector machine (SVM). The general methods for setting parameters include Grid search method (GS) and some typical swarm intelligence algorithms. However, these SVM variants only consider the margin but ignore the radius. This paper develops a new radius-margin-based SVM model with fruit fly optimization algorithm (FOA) called FOA-F-SVM, which considers the maximization of margin and the minimization of radius information. The FOA is adopted to optimize the penalty factor and parameter of RBF in F-SVM. The established model is solved in three steps, including initialization of matrix, decision of hyperplane and solution of transformation matrix. The effectiveness of the proposed FOA-F-SVM is evaluated against eight UCI datasets and eight comparison algorithms. The performance of the FOA-F-SVM is validated using the experimental results, and it is observed that FOA-F-SVM algorithm can produce more appropriate model parameters and significantly reduce the computational cost, which generates a high classification accuracy.