A Chaotic Elephant Herding Optimization Algorithm for Multilayer Perceptron based on Opposition-based Learning

This work proposes a new multilayer perceptron (MLP) training algorithm known as chaotic elephant herding optimization based on opposition-based learning (OCEHO). Two chaotic maps were incorporated into various phases of the standard EHO, as well as an opposition-based learning concept, to improve the global convergence of the population members and also to prevent the algorithm from falling into local optimal solutions. The proposed OCEHO-MLP model was then applied to two separate customer churn data-sets that are used for prediction, which were acquired from GitHub. The proposed model results were compared with those of the standard EHO and PSO used as MLP base trainers, as well as those of conventional MLP based on back-propagation. It is shown that the suggested training method OCEHO is significantly beneficial in enhancing the prediction rate of MLPs, with improved accuracy, convergence, and stability achieved in a fair amount of time.