The development of clusterwise regression model on gamma-normal mixed distribution with genetic algorithm

Abstract

Clusterwise regression is a statistical technique that combines clustering process and regression analysis. Cluster optimization was carried out using genetic algorithm (GA). GA is an optimization technique that adapts the theory of natural evolution starting from the formation of the initial population to produce the best generation. GA provides an optimal fitness value that describes the optimal clustering. This study aims to construct a clusterwise regression algorithm on Gamma-Normal mixed distribution using GA. Simulations were carried out to evaluate the results of the GA construction by generating data for various distributions. The simulation results on the Gamma distribution give an accuracy value 85% with cluster proportion 37% : 67%. Normal distribution with cluster proportion 51% : 49% yielded 95% accuracy. The highest accuracy is 98% in mixed distribution with cluster proportion 52% : 48%. Based on high accuracy value of the simulation results, it indicates the construction of an appropriate algorithm. This indicates that clusterwise regression Gamma-Normal mixed distribution with GA is able to cluster and model well.