A Comprehensive Analysis for P2P Lending Credit Risk Assessment on Imbalanced Data and Feature Selection

In the field of P2P lending, the problem of data imbalance has led to poor results of traditional credit risk assessment models. Addressing the issue of imbalanced data leading to ineffective classification is a popular research topic in data mining and machine learning. However, existing research has mainly focused on treating data imbalance to improve model performance, while largely overlooking the importance of data feature completeness. There is currently no systematic theoretical support for the relationship between imbalanced data and feature selection. To address this gap, our paper aims to investigate the impact of data feature completeness on imbalanced data in P2P lending. For this purpose, we selected 1,345,310 observations from Lending Club 2007-2018 Q4 as an empirical sample. We used LR, XGBoost, and DT as base classifiers and XGBoost as a tool for feature selection. Imbalanced learning methods were employed for evaluation. Our findings reveal that datasets with more complete features yielded better evaluation results when compared to the incomplete datasets used in existing studies. Moreover, we found that the Easy Ensemble-XGBoost approach showed the best results in solving the imbalance issue. Furthermore, we delved into the dynamics of imbalanced ratio and feature selection. Ultimately, by employing feature selection importance ranking, we obtained a more complete set of optimal features, thus enhancing the credit risk assessment model's performance in P2P lending