| In recent years,the national consumption level has gradually increased,and personal credit business has shown an explosive trend.The credit has provided convenience for residents and brought profits to financial institutions.There have been many related studies,but most of them use single-source data for analysis,and there are few studies on fusion of multiple datasets.In the era of big data,the information value and economic value of a single feature are not enough,and it is necessary to use a whole dataset to extract sufficient information.Therefore,this paper carries out the multiple datasets feature engineering based on the business sense,and explores the factors that affect the default risk in the customer’s loan application information,credit bureau information and historical behavior information.In the empirical analysis,the application information data is processed according to the business logic,including processing missing data and cleaning outliers,constructing mean and ratio derivative features based on literature experience and extracting components through SPCA.Then,this paper constructs derived features by the characteristics of different datasets and establishes XGBoost and LightGBM default risk prediction models respectively.The changes in evaluation indicators and the model-based feature importance index verify the necessity of multiple datasets and the effectiveness of feature engineering.Finally,this paper selects the LightGBM model and cross-validation method for parameter tuning and feature selection.Those works ensure the expandability of the multiple datasets processing and integration.The model has better robustness and better prediction performance through model fusion method.The data processing and modeling method provides reference for the recognition and identification of the default risk in the Internet personal loan,and help to minimize the insufficient credit problems caused by "information asymmetry",and improve the scientific nature of pre-loan approval for loans.At the same time,a series of suggestions for establishing a credit score system will also help the Internet finance company create a virtuous circle of business development. |
