The misclassification of loan applicants by credit scoring model is one of the main factors causing the loss of investors’ profits in P2P lending. Class imbalance of credit data is a main factor that affects classification performance of the model. Most existing methods of addressing class imbalance in credit scoring worked on improving the prediction accuracy for minority class samples (bad credit), which usually led to decreasing the prediction performance for majority class samples (good credit) significantly. In this paper, we propose a novel resampling ensemble model based on data distribution (REMDD) for imbalanced credit risk evaluation in P2P lending. REMMD solves class imbalance problem by using proposed undersampling method based on majority class data distribution (UMCDD). To further improve classification performance of REMMD, base classifiers with better comprehensive performance on the validation set are used to conduct class prediction. We validate the classification performance of REMDD on the three real and representative P2P lending credit datasets. The experimental results demonstrate that REMDD not only has good prediction performance for both majority class and minority class, but also effectively improves the comprehensive classification performance for imbalanced credit risk evaluation in P2P lending, compared with existing models.

信用评分模型对贷款申请人的错误分类是导致P2P借贷中投资者利润损失的主要因素之一。信用数据的类不平衡是影响模型分类性能的主要因素。解决信用评分中的类别不平衡的大多数现有方法都在提高少数群体样本（不良信用）的预测准确性，这通常会导致大大降低多数类别样本（良好信用）的预测性能。在本文中，我们提出了一种基于数据分布（REMDD）的新型重采样集成模型，用于P2P借贷中的不平衡信用风险评估。REMMD通过使用基于多数类别数据分布（UMCDD）的欠采样方法解决类别不平衡问题。为了进一步提高REMMD的分类性能，在验证集上具有更好综合性能的基本分类器用于进行类预测。我们在三个真实且具有代表性的P2P贷款信用数据集上验证REMDD的分类性能。实验结果表明，与现有模型相比，REMDD不仅对多数类和少数类都具有良好的预测性能，而且还有效地提高了P2P借贷中不平衡信用风险评估的综合分类性能。