Credit Default Prediction on Time-Series Behavioral Data Using Ensemble Models

Over the past few decades, credit default prediction has been central to managing risk in a consumer lending business. Credit default prediction allows lenders to optimize lending decisions, which leads to a better customer experience and sound business economics. Current models exist to help manage risk, but there is still exists space for better models that can outperform those currently in use. In this paper, we proposed a solution for the credit default prediction at double anonymized information including customer profile information and time-series behavioral data. Specifically, at the industrial-scale dataset provided by the credit default prediction competition of American Express, we leverage it to build a machine learning model that challenges the current model in application. Through the analysis of double anonymized information, we design an effective data processing flow, analyze the impact of time-series behavioral data, and derive useful latent features through feature augmentation and feature engineering, which ends up with a multi-model hybrid prediction integration scheme. The model consists of three modules: LightGBM, XGBoost, and Local-Ensemble, We use different feature combinations and individualized prediction schemes for each model to achieve efficient learning. Finally, the multi-model prediction results are ensemble to output the final result of a score-driven ensemble strategy. Experiments show that the method proposed in this paper has obvious advantages in solving the credit default prediction problem. We validate our model in the credit default prediction competition of American Express by ranking Top 10 of 4,874 teams.