Scholarlink

Credit risk assessment plays a crucial role in the banking and financial sector, ensuring financial stability and enabling well-informed lending decisions. With the increasing complexity of financial data, traditional statistical models and standalone machine learning (ML) techniques often fall short in predictive accuracy, adaptability, and interpretability. Recent developments in machine learning have significantly improved credit risk predictions, allowing financial firms to make better decisions. This work proposes an advanced credit risk forecasting model that combines XGBoost, CatBoost, and LightGBM for improved accuracy and dependability. The UCI Credit Card dataset, which contains important demographic and financial characteristics, is used to train the model for increased predicting confidence. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is used to improve discriminating between defaulters and non-defaulters. Additionally, grid search and cross-validation are used to adjust hyperparameters, which improves model performance. SHapley Additive exPlanations (SHAP) values are also used for improved interpretability, and it was discovered that predictors such as credit limitations, bill amounts, and past payment patterns were highly significant. According to experimental data, the proposed Hybrid Model (CatBoost + LightGBM) outperforms individual models like XGBoost (87%) and CatBoost (83%). Its best accuracy is 88.04%, precision is 91.77%, recall is 83.62%, F1-score is 87.51%, and AUC is 94.54%. This study adds to explainable AI in financial risk estimation by improving the model's interpretability and transparency. The findings support strong AI-driven financial modeling and allow financial institutions to make prudent, well-informed lending choices while adhering to legal requirements.

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

Hybrid Machine Learning Models for Credit Risk Prediction: An Explainable AI Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shivam Krishna,
  • Arun Solanki

摘要

Credit risk assessment plays a crucial role in the banking and financial sector, ensuring financial stability and enabling well-informed lending decisions. With the increasing complexity of financial data, traditional statistical models and standalone machine learning (ML) techniques often fall short in predictive accuracy, adaptability, and interpretability. Recent developments in machine learning have significantly improved credit risk predictions, allowing financial firms to make better decisions. This work proposes an advanced credit risk forecasting model that combines XGBoost, CatBoost, and LightGBM for improved accuracy and dependability. The UCI Credit Card dataset, which contains important demographic and financial characteristics, is used to train the model for increased predicting confidence. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is used to improve discriminating between defaulters and non-defaulters. Additionally, grid search and cross-validation are used to adjust hyperparameters, which improves model performance. SHapley Additive exPlanations (SHAP) values are also used for improved interpretability, and it was discovered that predictors such as credit limitations, bill amounts, and past payment patterns were highly significant. According to experimental data, the proposed Hybrid Model (CatBoost + LightGBM) outperforms individual models like XGBoost (87%) and CatBoost (83%). Its best accuracy is 88.04%, precision is 91.77%, recall is 83.62%, F1-score is 87.51%, and AUC is 94.54%. This study adds to explainable AI in financial risk estimation by improving the model's interpretability and transparency. The findings support strong AI-driven financial modeling and allow financial institutions to make prudent, well-informed lending choices while adhering to legal requirements.