WT combined early warning model and applications for loaning platform customers default prediction in smart city

Abstract

Due to the huge losses caused by the default of credit customers, the major loan platforms attach great importance to the testing and forecasting of non-performing loans. From the point of view of loan customer type identification and loan default, this paper constructs a WT early warning model of loan default client based on C5.0 decision tree, CART decision tree and CHAID decision tree in smart city. Aiming at the data characteristics of loan platform, this paper designs a posteriori combination mechanism of three algorithms and proposes a sub-model combined weighting algorithm based on the data characteristics of loan platform. Through the design performance test indicators, including sensitivity, accuracy, warning rate. The false alarm rate tests the performance of the combined and sub-model. In reality, using the real loan transaction dataset from the website of PPDAI to construct the WT model, it is found that the WT model overcomes the shortcomings of sub-model alone and achieves effective early warning of customer default. The empirical results show that the alarm rate of C5.0 is 29.17%, and the false alarm rate is 25.58%; the alarm rate of CART is 22.92%, and the false alarm rate is 20.59%; the alarm rate of CHAID is 23.75%, and the false alarm rate is 14.71%; the alarm rate of WT is 26.67%, and the false alarm rate is 17.65%; compared with other three algorithms, WT model is more effective in early warning of loan default.