The Best Thinking in Business Analytics from the Decision Sciences Institute

5. Improving Credit Scoring Accuracy via Sample Selection

—Rudy Setiono, National University of Singapore

Abstract

We present an approach for sample selection using an ensemble of neural networks for credit scoring. The ensemble determines samples that can be considered outliers by checking the prediction accuracy of the neural networks on the original training data samples. Those samples that are consistently misclassified by the neural networks in the ensemble are removed from the training dataset. The remaining data samples are used to train another neural network for rule extraction. Our experimental results on a publicly available benchmark dataset show that by eliminating the outliers, neural networks can be trained to achieve better predictive ...