A decision hyper plane heuristic based artificial immune network classification algorithm
来源期刊:中南大学学报(英文版)2013年第7期
论文作者:DENG Ze-lin(邓泽林) TAN Guan-zheng(谭冠政) HE Pei(何锫) YE Ji-xiang(叶吉祥)
文章页码:1852 - 1860
Key words:artificial immune network; decision hyper plane; recognition ability; classification
Abstract: Most of the developed immune based classifiers generate antibodies randomly, which has negative effect on the classification performance. In order to guide the antibody generation effectively, a decision hyper plane heuristic based artificial immune network classification algorithm (DHPAINC) is proposed. DHPAINC taboos the inner regions of the class domain, thus, the antibody generation is limited near the class domain boundary. Then, the antibodies are evaluated by their recognition abilities, and the antibodies of low recognition abilities are removed to avoid over-fitting. Finally, the high quality antibodies tend to be stable in the immune network. The algorithm was applied to two simulated datasets classification, and the results show that the decision hyper planes determined by the antibodies fit the class domain boundaries well. Moreover, the algorithm was applied to UCI datasets classification and emotional speech recognition, and the results show that the algorithm has good performance, which means that DHPAINC is a promising classifier.
DENG Ze-lin(邓泽林)1, 2, TAN Guan-zheng(谭冠政)1, HE Pei(何锫)2, YE Ji-xiang(叶吉祥)1, 2
(1. School of Information Science and Engineering, Central South University, Changsha 410083, China;
2. School of Computer and Communication Engineering, Changsha University of Science and Technology,
Changsha 410076, China)
Abstract:Most of the developed immune based classifiers generate antibodies randomly, which has negative effect on the classification performance. In order to guide the antibody generation effectively, a decision hyper plane heuristic based artificial immune network classification algorithm (DHPAINC) is proposed. DHPAINC taboos the inner regions of the class domain, thus, the antibody generation is limited near the class domain boundary. Then, the antibodies are evaluated by their recognition abilities, and the antibodies of low recognition abilities are removed to avoid over-fitting. Finally, the high quality antibodies tend to be stable in the immune network. The algorithm was applied to two simulated datasets classification, and the results show that the decision hyper planes determined by the antibodies fit the class domain boundaries well. Moreover, the algorithm was applied to UCI datasets classification and emotional speech recognition, and the results show that the algorithm has good performance, which means that DHPAINC is a promising classifier.
Key words:artificial immune network; decision hyper plane; recognition ability; classification