Evaluation of Genetic Algorithms for tuning SVM parameters in multi-class problems

Support Vector Machine (SVM) is a useful technique for data classification with successful applications in different fields of bioinformatics, image segmentation, data mining, etc. A key problem of these methods is how to choose an optimal kernel and how to optimize its parameters in the learning process of SVM. The objective of this study is to propose a Genetic Algorithm approach for parameter optimization to solve this kind of problem. The proposed method is compared with grid algorithm, a traditional method for parameter setting, by conducting some experiments using different benchmark data sets. The results observed show better performance of hybrid GA-SVM method by improving classification accuracy.