Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
Double-layer Adaptive Stacking Credit Risk Prediction Based on mRMR-ADASYN
1School of Science, Changchun University, China.
2Graduate School, Changchun University, China.
ABSTRACT
This paper proposes a two-layer adaptive Stacking model based on mRMR-ADASYN to address the challenges of high-dimensional credit data and severe class imbalance between positive and negative samples. In the first stage of the model, the variable features are screened through the combination of mRMR-ADASYN, and the positive samples are oversampled. This paper uses the fusion evaluation index TotalScore to screen the five bestperforming base learners in the second stage. It uses Bayesian optimization to adjust parameters as the first layer of the Stacking model. In the third stage, the model adaptively selects the best meta-learner for fusion among Logistic Regression (LR), Decision Tree (DT), and Extreme Gradient Boosting (XGBoost) as the second layer of the Stacking model. It uses the Snake Optimization Algorithm (SO) for parameter optimization. The experimental results show that the classification performance of the XGBoost meta-learner fusion model tuned by the SO is better than other fusion methods.
Keywords: mRMR, ADASY, Snake optimization algorithm, Credit risk, Stacking, XGBoost.
Download PDF
1School of Science, Changchun University, China.
2Graduate School, Changchun University, China.
ABSTRACT
This paper proposes a two-layer adaptive Stacking model based on mRMR-ADASYN to address the challenges of high-dimensional credit data and severe class imbalance between positive and negative samples. In the first stage of the model, the variable features are screened through the combination of mRMR-ADASYN, and the positive samples are oversampled. This paper uses the fusion evaluation index TotalScore to screen the five bestperforming base learners in the second stage. It uses Bayesian optimization to adjust parameters as the first layer of the Stacking model. In the third stage, the model adaptively selects the best meta-learner for fusion among Logistic Regression (LR), Decision Tree (DT), and Extreme Gradient Boosting (XGBoost) as the second layer of the Stacking model. It uses the Snake Optimization Algorithm (SO) for parameter optimization. The experimental results show that the classification performance of the XGBoost meta-learner fusion model tuned by the SO is better than other fusion methods.
Keywords: mRMR, ADASY, Snake optimization algorithm, Credit risk, Stacking, XGBoost.
Download PDF