Bankruptcy Prediction by Genetic Ant Colony Algorithm

Abstract

Corporate bankruptcy is a hot topic in economical research. Traditional methods cannot reach satisfying classification accuracy due to the high dimensional features. In this study, we proposed a novel method based on wrapper-based feature selection. Moreover, a novel genetic ant colony algorithm (GACA) was proposed as the search method, and the rule-based model was employed as the classifier. Stratified K-fold cross validation method was taken as the statistical resampling to reduce overfitting. Simulations take 1,000 runs of each algorithm on the dataset of 800 corporations during the period 2006-2008. The results of the training subset show that the GACA obtains 84.3% success rate, while GA obtains only 48.8% and ACA obtains 22.1% success rate. The results on test subset demonstrate that the mean misclassification error of GACA is only 7.79%, less than those of GA (19.31%) and ACA (23.89%). The average computation time of GACA is only 0.564s compared to the GA (1.203s) and ACA (1.109s).