WT Model & Applications in Loan Platform Customer Default Prediction Based on Decision Tree Algorithms
Due to the huge losses caused by the bad credit customers, loan platforms attach great attention to testing and forecasting of bad loans. From perspectives of both loan customer type identification and loan default prediction, we initially constructed a WT early warning model for loan default client prewarning based on C5.0 decision tree, CART decision tree and CHAID decision tree in this paper. WT model is set with weighted calculating algorithms. Considering the data characteristics of loan platform, we designed a posteriori combination algorithm of three sub-models: C5.0, CART, CHAID, and performance test indicators: sensitivity, accuracy, warning rate, false alarm rate. In empirical research, we used the real loan transaction dataset of a bank in Taiwan to construct the WT model of the bank, and found that WT model overcomes the shortcomings of each sub-model respectively and achieves effective prewarning of customer default. The experimental results show that the alarm rate of test data set is 26.93% and the false alarm rate is 18.33% and the accuracy rate is 81.67% when applying the WT model. Loan platforms can acquire both high customer default prediction accuracy and high alarm rate by applying WT prewarning model. Both the research method and experiment results in this paper are meaningful to loan platform operation.
KeywordsWT model Decision tree algorithms Customer default prewarning Loan platforms
The paper is supported by Natural Science Foundation of China (Grants No. 91646112); The Key Programs of Science and Technology Department of Guangdong Province (Grants No. 2016A020224001).
- 1.Bache, K., Lichman, M.: UCI Machine Learning Repository (2013)Google Scholar
- 4.Ge, P., Peng, M.: The optimization and application of the decision tree algorithm in the classification prediction 47, 936–939 (2016)Google Scholar
- 5.Imtiaz, S., Brimicombe, A.J.: A better comparison summary of credit scoring classification. Int. J. Adv. Comput. Sci. Appl. 8, 1–4 (2017)Google Scholar