Improvement of Neural Networks with Swarm Intelligence Algorithm for Credit Scoring

Abstract

Neural networks are widely used in automatic credit scoring systems with high accuracy and outstanding efficiency. However, in the absence of prior knowledge, it is difficult to determine the combination of parameters, which makes its application limited in practice. This paper presents a higher accurate and robust credit scoring model based on neural networks that have been trained with the optimal swarm intelligence algorithm. Specifically, we trained neural network with seven different swarm intelligence algorithm (bat algorithm, chicken swarm optimization, cuckoo search optimization, firefly algorithm, particle swarm optimization, social spider algorithm, and whale swarm algorithm) to find out the superior combination of parameters in the neural network and to identify the swarm intelligence algorithm seeking the superior solution most efficiency. It shows that the neural networks trained with swarm intelligence algorithm outperforms competing models (logistic regression, naive Bayesian, determinant analysis, K nearest neighbor, decision tree, and support vector machine), inter alia, the neural network trained with social spider algorithm performs the best. Better performance of the neural network is particularly salient with larger dataset, thus making it amenable for real-time implementation.

Keywords: credit scoring, neural network, swarm intelligence algorithm

JEL Classification: G21