Enhancing Credit Scoring through a Multi-Stage Hybrid Ensemble Classification Model with Feature Engineering and Data Balancing

Financial institutions face significant credit risk when evaluating credit applications, making fraud detection and prevention a primary task for all financial institutions and service companies. After decades of research and development, the credit scoring model has been improved by artificial intelligence and machine learning. This research proposes a new multi-stage hybrid ensemble classification model for improving the efficiency of credit scoring applications by leveraging artificial intelligence and machine learning advancements. A new weighted ensemble filter and an improved borderline SMOTE for minority oversampling are proposed to enhance the dataset quality by identifying the most suitable set of features for analysis, addressing the bias caused by an imbalanced class distribution, and improving predictive accuracy. Moreover, to reduce the variance and enhance the accuracy, a new nested ensemble classification model was introduced to enhance the predictive accuracy of credit scoring performance. The proposed model outperformed four credit scoring datasets, achieving an improved accuracy and AuC of 92.43% and 0.968 for the Australian credit dataset, 83.16% and 0.901 for the German credit dataset, and an improved AUC of 0.938 for the Japanese credit dataset. The experimental findings validate the practical efficiency of the proposed model, which outperforms competing models owing to its unique combination of feature engineering, balancing techniques, and use of a novel nested ensemble classifier. The study offers financial institutions not only a more robust tool for credit scoring to make more reliable credit decisions but also reduces operational costs associated with manual evaluation processes, risk management, and fraud mitigation.