Enhanced support vector machine based on metaheuristic search for multiclass problems

Machine learning is an important field of artificial intelligent researches and it highly growing for real intelligent applications systems that relate brain computer interface to human brain activities. Support Vector Machine (SVM) is a popular machine learning approach, which can be used for pattern recognition, prediction and classification with many diverse applications. However, the SVM has many parameters, which have significant influences on the performance of SVM in terms of its prediction accuracy that is very important measure specifically with critical applications such that used in Medical applications. This paper proposed an enhanced SVM, which employs a meta-heuristic method, called scatter search to determine the optimal or near optimal values of the SVM parameters and its kernel parameters in multi-classification problem. Scatter search has the potential to determine the appropriate values of parameters for machine learning algorithms due its flexibility and sophistication. Therefore, the proposed method integrates the advantages of scatter search method with SVM to specify the appropriate setting of SVM parameters. The experimental results on lung cancer datasets and other standard datasets prove that the scatter search is practical method for tuning SVM parameters and enhance its performance, where the achieved results are better and comparable to other related methods.