A TWO-STEP SUPERVISED LEARNING ARTIFICIAL NEURAL NETWORK FOR IMBALANCED DATASET PROBLEMS

In this paper, a two-step supervised learning algorithm of a single layer feedforward Artificial Neural Network (ANN) is proposed for solving Unbalanced dataset problems. Levenberg Marquart backpropagation learning algorithm is utilized in the first step learning, while the second step learning mechanism is introduced by optimizing the decision threshold of the step function at the output layer of ANN using particle swarm optimization (PSO). After all the steps learning are accomplished, the best weights and decision threshold value are obtained to be used for testing process. Several unbalanced datasets, which are available in UCI Machine Learning Repository, are chosen as case study. The prediction performance is assessed by Geometric Mean (G-mean), which is a standard measure to indicate the efficiency of classifier for Unbalanced datasets. Based on the experimental results, the proposed method is able to provide good G-mean value compared with the conventional ANN approaches.