A Boosting Approach Based on Bat Optimization in MLP Neural Networks: Classification Task

Achieving optimal weights of artificial neural networks (ANN) is crucial issue. Some approaches have attempted to obtain efficient weights using meta-heuristic algorithms. In this paper, we utilize bat optimization algorithm and its modifications to gain optimal weights of multilayer perceptron neural networks (MLP) applied in adaboost algorithm as weak classifiers. Different modifications of bat algorithm improve the exploration and exploitation capability to optimize (minimize) mean square error (MSE) of MLP. Bat algorithm is based on the echolocation behavior of real bats and it derives some benefits of population based and local search algorithms. Experimental results illustrate that the proposed hybrid approach improves the detection rate of MLP neural networks for classification task. Also, Wilcoxon test shows that this approach is more efficient than some similar base researches in this problems.