A Modified Artificial Neural Network Learning Algorithm for Imbalanced Data Set Problem

Abstract

A modified learning algorithm of Artificial Neural Networks (ANN) is introduced in this paper to solve imbalanced data set problems. In solving imbalanced data set, it is critical to predict the minority class due to their imbalanced nature. In order to improve the standard ANN classifier prediction performance, this paper focuses on optimizing the decision boundary of the step function at the output layer of ANN using particle swarm optimization (PSO). A feedforward ANN is chosen in this study. Firstly, a conventional back propagation algorithm is employed to train the ANN. PSO is then applied to train the real predicted output of training data from this trained network. As the result, the optimum value of decision boundary is found and applied to the classifier. Prediction performance is assessed by G-mean, which is a measure to indicate the efficiency of classifiers for imbalanced data sets. Based on experimental results, the proposed model is able to solve imbalanced data sets problem with better performance compared to the standard ANN.