The Prediction Of Personal Loan Default Based On LightGBM Model

With the rapid development of Internet lending on big data platforms and the rapid expansion of the scale of online loans,the problems of imprudent risk management and inadequate monitoring of fund usage have become more and more serious,especially,the extremely high bad debts make the lending platform operators and investors pay a huge price.Therefore,establishing an efficient loan business default risk identification model to guarantee the healthy and long-term development of lending platforms and the Internet finance industry is a research work of great practical significance.However,the existing researches for loan default models are dominated by traditional machine learning methods such as logistic regression and decision trees,which are ineffective for multidimensional and massive data.In this paper,we leverage LightGBM algorithm to overcome such shortcoming and provide a new research idea for loan default model in multidimensional and massive data.The analyses of a real data from Aliyun Tianchi Big Data Competition Platform are conducted to access the performance of the proposed method,including data pre-processing,feature engineering,model training and prediction.As compared to several traditional machine learning models,the LightGBM algorithm not only guarantees the efficiency of classification but also summarizes the factors that affect borrowers’ default through feature importance ranking,which provides reference value for the risk management of lending platforms.