Research on Personal Credit Evaluation Based on Bayesian Optimization and Fusion Models

金融科技的迅速发展推动个人信贷市场规模扩大，信用风险评估成为金融机构风险管控的核心环节。针对这一问题，本研究提出了一种基于贝叶斯优化和Blending融合模型的信用评估方法。通过数据预处理和特征选择构建高质量训练数据基础，超参数优化使用贝叶斯优化方法构建模型。为了评估模型性能，选取Logistic回归、随机森林、LightGBM和XGBoost模型进行了对比实验，其中Blending融合模型在所有模型中表现最佳。实验结果表明，本文提出的 Blending融合模型在个人信用风险评估中具有明显优势，能够有效整合不同模型的特点，提供更精准的信用风险预测。此外，贝叶斯优化在调参效率与模型性能上均优于网格搜索与随机搜索，其全局寻优特性尤其适用于高维数据场景。The rapid development of financial technology drives the expansion of personal credit market scale, and credit risk assessment becomes the core link of risk control of financial institutions. To address this problem, this study proposes a credit assessment method based on Bayesian optimization and Blending fusion model. The high-quality training data base is constructed through data preprocessing and feature selection, and the hyperparameter optimization uses Bayesian optimization to construct the model. In order to evaluate the model performance, Logistic regression, random forest, LightGBM and XGBoost models are selected for comparative experiments, in which the Blending fusion model performs the best among all models. The experimental results show that the Blending fusion model proposed in this paper has obvious advantages in personal credit risk assessment, and can effectively integrate the features of different models to provide more accurate credit risk prediction. In addition, Bayesian optimization outperforms grid search and random search in terms of tuning efficiency and model performance, and its global optimization characteristics are especially suitable for high-dimensional data scenarios.