Credit Risk Prediction in an Imbalanced Social Lending Environment Based on XGBoost

The credit data of small enterprises are imbalanced, which leads to the failure of the general classification models in predicting the credit performance directly. In order to reduce the negative impact of imbalanced data set on model, we construct a credit rating model based on K-XGBoost from the perspective of a hybrid model, which is a new hybrid model proposed by this paper combining the advantages of both KMeans++ and XGBoost. And we construct XGBoost credit rating models based on SMOTE series algorithms from the perspective of random oversampling. According to empirical analysis, the indicators of K-XGBoost model are superior to the XGBoost model of dataset balanced by the Borderline SMOTE algorithm, and can reduce the unnecessary costs by accelerating modeling.In order to meet the needs of commercial banks for risk control and market share, we construct a three-class credit rating model based on XGBoost, which divides customers into stable good customers, unstable good customers and bad customers. When the loan policy is tightened, only stable customers will be the target customer group. Empirical research shows that the above three models have good performance and strong generalization ability.