A Novel Multi-Reduced Support Vector Machine

For solving the problem that support vector machine (SVM) is restricted to work well on the small sample sets, a novel multi-reduced support vector machine (MRSVM) is proposed. MRSVM has two reduced steps. Firstly, a novel reduced support vector machine on density clustering (RSVM-DC) is proposed. This algorithm focuses on dealing with a sample set through density clustering prior to classifying the samples. After clustering the positive samples and negative samples, the algorithm picks out such samples that locate on the edge of clusters as reduced samples. These reduced samples are treated as the new training sample set used in SVM´s classifier system. Additionally, it can also improve precision by reducing the percentage of counterexamples in kernel object epsi-area. Secondly, a novel reduced algorithm in high dimension through kernel mapping is proposed. Based on the linear classified character in high dimension, it can select the edge samples of each kind sample set to reduce the kernel computing times. Experiment results show that not only efficiency but also classification precision are improved by MRSVM, compared with other reduced algorithms