Hybrid fuzzy support vector regression analysis

Abstract

Support Vector Machines (SVM) has recently received a great deal of attention in regression and classification problems for crisp data. However, the fuzzy structure of the system should be considered if available information is uncertain or imprecise. In this paper, a new approach called Hybrid Fuzzy Support Vector Regression (HF-SVR) is introduced for the linear and non-linear fuzzy regression modeling. According to the proposed algorithm, parameter estimates are obtained from the solutions of two optimization problems by using the basic idea underlying SVM and the least squares principle. Furthermore, different learning machines for non-linear fuzzy regression can be constructed according to the selection of kernel function. In order to compare HF-SVR with the previously published fuzzy support vector regression methods, the data sets used in these papers are adopted. Based on the findings obtained from the numerical applications, it is determined that proposed method gives remarkable results according to the measure of mean square error.