Differential Evolution Based Optimization of SVM Parameters for Meta Classifier Design

In this paper, we have devised a meta classifier model by simultaneously optimizing different evaluation criteria of classifier performance. For this purpose, a support vector machine (SVM) is used as the underlying classifier and its ernel parameters are optimized using differential evolution. We have also formulated a new fitness function combining ifferent classifier evaluation criteria, i.e., accuracy, sensitivity and specificity. The performance of the proposed meta classification approach is demonstrated to be superior to those of the individual classifiers and also several other meta classifiers based on analyses on three real-life datasets.