Parameter Tuning of MLP, RBF, and ANFIS Models Using Genetic Algorithm in Modeling and Classification Applications

Nowadays, soft computing algorithms are used to simulate and solve complex problems in different fields. Each of the existing algorithms requires the proper adjustment of parameters to produce the outputs tailored to the user's needs. Using trial-and-error methods to find the correct parameter values is time-consuming and does not guarantee the optimum value. In this paper, a genetic algorithm is employed to obtain optimal parameter values for multi-layer perceptrons, radial basis function, and adaptive neuro-fuzzy inference system algorithms. For each of these algorithms, different coding for chromosomes is provided according to the user's needs. In this approach, the network architecture also changes, unlike many existing schemes that initially assume the same topology for all algorithms. Simulations on standard datasets, including real and artificial datasets in the field of modeling and classification, are presented to verify the resulting performance.