Multiple Optimized Ensemble Learning for High-Dimensional Imbalanced Credit Scoring Datasets

Abstract Banks determine the financial credibility or the credit score of the applicants before allocating loans to them. In recent decades, several machine learning algorithms have been developed to automate the decision-making process by constructing an effective credit scoring models. However, the high-dimensional and imbalanced credit datasets significantly degrade the models' classification ability. In this study to overcome these issues, a novel multiple-optimized ensemble learning (MOEL) is proposed to build a reliable and accurate credit scoring model. MOEL, first generates multiple diverse optimized subsets from various weighted random forests (WRFs), and from each subset more effective and relevant features are selected. A new evaluation measure is then applied to each subset to determine which subsets are more effectively optimized for the ensemble learning process. The subsets are then applied to a novel oversampling strategy to provide balanced subsets for the base classifier, which lessens the detrimental effects of imbalanced datasets. Finally, to further improve the performance of the base classifier, a stacking-based ensemble method is applied to the balanced subsets. Six credit-scoring datasets were used to evaluate the model's efficacy using the F1 score and G-mean metrics. The empirical results on these datasets demonstrate that MOEL achieves the best value of F1_score and G-mean with a mean ranking of 1.5 and 1.333, respectively.