A New Method of Sample Reduction for Support Vector Classification

Abstract

As a powerful tool in machine learning, Support Vector Machine(SVM) also suffers from expensive computational cost in the training phase due to the large number of original training samples. To overcome this problem, this paper presents a new method based on a two steps of sample reduction to reduce training samples. This algorithm includes cluster detection by Fuzzy C-Means Clustering (FCM) Cluster and sample reduction by Multivariate Gaussian Distribution (MGD). In its implementation, the FCM algorithm is used to cluster the original training samples, and then the MGD is used to reduce the training samples by choosing the only boundary samples for the next training. Experiments show that the algorithm accelerates the training speed without the decrease of classification accuracy.