A kernel-based fuzzy greedy multiple hyperspheres covering algorithm for pattern classification

This paper presents a kernel-based fuzzy greedy multiple hyperspheres covering algorithm for pattern classification. In the training process all training data of each class are covered by multiple hyperspheres constructed, each of which encompasses as many data as possible via a greedy method. In the classification process a fuzzy membership function is defined to label the testing samples. Furthermore, we introduce kernel methods into the proposed method. To investigate the effectiveness of our approach, experiments are done on artificial data sets and six real data sets. Experimental results show that our algorithm not only can acquire the lower time complexity in training and the better classification accuracies than two hyperspheres-based classification methods, but also can achieve the comparable performance to the classical support vector machines.