基于贝叶斯优化LightGBM的个人信用评估模型

Personal Credit Evaluation Model Based on Bayesian Optimization of LightGBM

针对传统信用评估模型无法处理大规模不平衡数据以及训练用时长、评估不准确等问题,提出了一种优化的个人信用评估模型,模型基于梯度提升框架LightGBM,结合贝叶斯全局优化算法进行个人信用评估。为验证模型适用性,采用Lending Club公开数据集进行相关实验,并与逻辑回归、随机森林、XGBoost模型的预测结果进行比较。实验结果表明,该模型的个人信用评估效果更好,评估准确率达到99.97%,少数类样本F1-score达到89.02%。

Aiming at the problems that traditional credit evaluation models cannot handle large-scale imbalanced data,train-ing time,and inaccurate evaluations,an optimized personal credit evaluation model is proposed.The model is based on the gradient boosting framework LightGBM,combined with the Bayesian global optimization algorithm for personal credit evaluation.In order to verify the applicability of the model,the Lending Club public data set is used to conduct related experiments and compared with the prediction results of logistic regression,random forest,and XGBoost models.The experimental results show that the personal credit evaluation effect of this model is better,the evaluation accuracy rate reaches 99.97%,and the F1-score of minority samples reaches 89.02%.