Fuzzy Nonlinear Regression Modeling With Radial Basis Function Networks

In this article, we extend the popular supervised learning technique radial basis function network (RBFN) for regression modeling based on fuzzy responses and exact predictors. For this purpose, we suggest a penalized squared error ridge-based method to estimate the model components including fuzzy parameters and exact tuning constants. The performance of the newly proposed model is examined via established goodness-of-fit criteria and the effectiveness is demonstrated within some numerical application examples. Following the obtained results it is indicative that the fuzzy RBFN regression model outperforms conventional fuzzy nonlinear and multiple regression models and provides more accurate results for nonlinear regression problems.