Scholarlink

Assessing credit risk and ensuring financial stability in lending institutions hinge on accurate credit default prediction. This study evaluates the efficacy of diverse machine learning models in predicting credit default using a dataset comprising varied financial attributes. Initially, we present the outcomes of five baseline models—random forest, XGBoost, kNN, logistic regression, and neural network—in terms of precision, recall, F1-score, and overall accuracy. We subsequently investigate the impact of principal component analysis (PCA) and hyperparameter tuning via GridSearchCV on model performance. Our results demonstrate notable enhancements in model performance, particularly in accurately identifying non-default instances, upon integrating PCA and hyperparameter optimization. Furthermore, we underscore the adeptness of the multilayer perceptron (MLP) neural network model, which features two hidden layers with 64 and 32 neurons respectively, in capturing intricate data patterns. This research contributes to advancing credit default prediction methods and underscores the significance of employing comprehensive modeling techniques to bolster predictive accuracy in financial risk assessment.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Credit Score Prediction Through Hybrid Machine Learning Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ruslan Murzagaliyev,
  • Bizhigit Sagidolla,
  • Shirali Kadyrov

摘要

Assessing credit risk and ensuring financial stability in lending institutions hinge on accurate credit default prediction. This study evaluates the efficacy of diverse machine learning models in predicting credit default using a dataset comprising varied financial attributes. Initially, we present the outcomes of five baseline models—random forest, XGBoost, kNN, logistic regression, and neural network—in terms of precision, recall, F1-score, and overall accuracy. We subsequently investigate the impact of principal component analysis (PCA) and hyperparameter tuning via GridSearchCV on model performance. Our results demonstrate notable enhancements in model performance, particularly in accurately identifying non-default instances, upon integrating PCA and hyperparameter optimization. Furthermore, we underscore the adeptness of the multilayer perceptron (MLP) neural network model, which features two hidden layers with 64 and 32 neurons respectively, in capturing intricate data patterns. This research contributes to advancing credit default prediction methods and underscores the significance of employing comprehensive modeling techniques to bolster predictive accuracy in financial risk assessment.