Default Prediction by Using Machine Learning Methods

Abstract

In this paper, we apply the machine learning methods to predict default of companies in Asian financial crisis using financial statements from 1994 to 1996. Logistic regression, neural network, and support vector machines are used to conduct the default prediction model. We use under-sampling technique and SMOTE(Synthetic Minority Over-sampling Technique) to solve the imbalanced dataset problem. Also, we compare the results with two subgroups of features, one group is from one financial statement prior to default, and the other is from three financial statements from 1994 to 1996. In addition, we discuss the performance of the machine learning methods by comparing the statistic measures.