Performance Assessment of Combination in Stacking Ensemble Model for Credit Default Classification

Brenda Sylviasyah
2020 International Journal of Advanced Trends in Computer Science and Engineering  
Credit Default is one of the most discussed and reviewed problems in a financial institution. The ever-changing factors and variables towards the consideration of credit grant remains a challenge to prevent loss caused by non-performing loans. In the light of the machine learning era, one of the methods that can be applied to solve this problem is by using classification models. Ensemble Method is known for improving better model performance. This paper would focus on assessing the performance
more » ... f various combination of 7 well-known classification algorithm such as SVC, Decision Tree (CART), Naïve Bayes, Logistic Regression, Random Forest, Extra Trees and XG-Boost.Out of using total 848combination of 1-7 algorithm with 7 meta classifier, this experiment shows that models from stacking ensemble does indeed generally perform better compared to single base-classifier. Assessing the performance between two credit datasets with different product type, the experiment concludes that the most ideal iteration is between 2-5 base-learner combination using SVC as the meta-classifier for this case. This study also suggests the usage of cost function for assessing credit classification problem for its ability to simulate a projection of loss and gain by implementation.
doi:10.30534/ijatcse/2020/194942020 fatcat:cwycalg4wjexbot5tw7jvavrei