Application of Xgboost Model to Predict Bank Loan

Nowadays the bank loan has been a high-frequent behavior in the field of finance. Predicting the deposits paid by the lenders can effectively reduce the risk of credit fraud and avoid the loss of bank. Traditional methods like classification regressons and support vector machines are difficult to deal with large sets of data and with relatively low accuracy. In this paper, a novel and robust method called XGBoost is applied in the bank loan and achieves a high accuracy. Also, proving the accessibility of XGBoost model in the business analysis field. The data set used in this study comes from a real bank customer database, which contains 16 variables. Regarding the categorical variables, this paper used one-hot encoding to make these variables can be identified by the model. Following by two steps feature selection, firstly used LVQ to filter variables with low importance, then used RFE to further select 11 variables. After that, this paper used these variables to build a XGBoost model to predict the bank loan deposit. Through hyperparameters optimization, this paper achieved a final accuracy of 80.62%, which shows an excellent model performance. In the future, this model can be used in the bank loan decision systems to determine the deposits that paid by lenders.