ROC Analysis by FCM Classifier with Particle Swarm Optimization

Abstract

Since different types of classifiers work best for different types of data, our approach is to parameterize the classifier and tailor them to individual data set. In the proposed fuzzy c-means (FCM) classifier, parameter values are chosen by evolutionary algorithms such as the particle swarm optimization (PSO). The free parameters of membership functions are optimized in order to increase true positive rate or true positive count. The golden section search is applied to select a cut-off point, which realizes a preselected positive count or false positive count. This makes classification decisions more pragmatic than those by the standard methods with missclassification rates or loss functions.