Fuzzy Support Vector Learning Algorithm for Mixed Attributes Data
Zhongdong Wu, Jianping Yu, Yanping Li, Weixin Xie · 2006
A new fuzzy support vector learning algorithm (FSVLA) for mixed attributes data is investigated by utilizing FSVM (Fuzzy Support Vector Machine), which was proposed previously. Firstly, the similarity degree between a pair of data with mixed attributes is defined. Then a kernel matrix based on mixed similarity degree is constructed and proved to be a Mercer kernel. So, the original mixed attributes space is mapped into a canonical high dimensional space with simplex continuous attributes with preserving the primary information in the given data. By learning algorithm of SVM with good generalization performance, the FSVLA was proposed, which has a small set of fuzzy if-then rules. The new learning algorithm has good generalization ability and linguistic interpretation, and the numerical experiments illustrate the effectiveness of the new learning algorithm.