The Optimal Rule Structure for Fuzzy Systems in Function Approximation by Hybrid Approach in Learning Process

A hybrid approach of learning process is investigated to optimize the fuzzy rule structure of the fuzzy system for function approximation. First, if-then rules are initialized more much than usual and then are optimized via deployment of a genetic algorithm. Subsequently, the supervised gradient descent algorithm (incorporated momentum technique) is utilized in order to tune the fuzzy rule parameters. Experimental results are presented that indicate significant improvement in term of accuracy in function approximation can be achieved during deployment of the Standard Additive Model (SAM) by adopting the hybrid approach.