Given the credit risk losses brought by Loan default to commercial banks, this paper, based on the America Express data set of the Kaggle platform, builds a hybrid model to predict customer default, to reduce credit risk. We first conduct data cleaning and feature engineering processing, dividing the data into category data and continuous data, and calculating the correlation between different features. We build DNN and lightGBM for credit default prediction. From the comprehensive results, the prediction effect of the mixed model is significantly better than that of the single prediction model. Among these models, our hybrid approach achieves the leading metric of 0.798. This metric is higher than XGBoost, LightGBM, and DNN by 0.013, 0.002, and 0.008 respectively.