International Journal of Computer Theory and Engineering

Abstract

The performance of the support vector machine (SVM) is determined to a great extent by the parameter selection. In order to improve the learning and generalization ability of SVM, in this paper, an improved fruit fly optimization algorithm (IFOA) was proposed to optimize kernel parameter and penalty factor of SVM. In IFOA, the fruit fly group is dynamically divided into advanced subgroup and drawback subgroup according to its own evolutionary level. A global search is made for the drawback subgroup under the guidance of the best individual and a finely local search is made for the advanced subgroup in which the fruit flies do Levy flight around the best individual. Two subgroups exchange information by updating the overall optimum and recombining the subgroups. Getting rid of local optimum and improve search ability are ensured by making those changes in basic FOA. The performance of the IFOA and classification accuracy of optimized SVM based on IFOA are respectively examined through several typical benchmark functions and classical data sets from UCI benchmark. The experiment results show that the performance of the new algorithm is obviously more successful than FOA and it is also an effective SVM parameter optimization method which has better performance than some other methods.