On Granular Rough Computing: Covering by Joint and Disjoint Granules in Epsilon Concept Dependent Granulation

In this work we present the optimization methods of epsilon concept-dependent granulation. We consider two cases of parallel covering and granulation, based on joint and disjoint granules. Additionally we check two variants of majority voting, the first one based on descriptors, which are epsilon-indiscernible with the centers of granules, and the second variant uses all descriptors of respective granules. We verify the effectiveness of our methods on the real data sets from UCI Repository using the SVM classifier. It turned out that disjoint granules versus joint give almost identical results of classification with a significant acceleration of the granulation process. Additionally, the majority voting, based on the epsilon indiscernible descriptors, stabilised the process of granulation in terms of the accuracy of classification. This is a significant result, which lets us to accelerate the process of classification for many popular classifiers at least for k-NN, Naive Bayes, many rough set methods and the SVM classifier, which is supported by our recent works.