Fuzzy Support Vector Machines Based on Fuzzy Similarity Degree

摘要

Support Vector Machines (SVM) for classification problem with fuzzy inputs is proposed. This is based on Mercer kernels, or equivalently, positive definite kernel matrix. We use fuzzy similarity degree as similarity (dissimilarity) between two fuzzy vectors, and construct positive definite kernel matrix that is based on fuzzy similarity degree. We call this novel SVM as Fuzzy Support Vector Machines (FSVM). The merit of FSVM is that it can incorporate with domain knowledge represented by fuzzy IF-THEN rules to improvement of performance of the conventional SVM in incomplete numeral data set for training. The simulation results are very encouraging.