A new hybrid ensemble model with voting-based outlier detection and balanced sampling for credit scoring. (15th July 2021)

• Record Type: Journal Article
• Authors: Zhang, Wenyu; Yang, Dongqi; Zhang, Shuai
• Is Part Of: Expert systems with applications. Volume 174(2021)
• Publication Date: 2021-07-15
• Subjects: Machine learning -- Ensemble modeling -- Outlier detection -- Balanced sampling -- Credit scoring; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2021.114744
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: A new hybrid ensemble model is proposed for credit scoring. The outlier adaptability is enhanced by the voting-based outlier detection method. The under-sampling method is extended to handle data imbalance. The stacking-based ensemble enhances the predictive power of the proposed model. The proposed model outperforms the benchmark ensemble models. Abstract: The credit scoring system has been revolutionized with the development of the financial system and has received increasing attention from the academia and industry. Artificial intelligence technology has reshaped credit scoring through predictive classification. In this study, a new hybrid ensemble model with voting-based outlier detection and balanced sampling is proposed to achieve superior predictive power for credit scoring. To avoid noise-filled data from misleading the classifier training, a new voting-based outlier detection method is proposed to enhance the classic outlier detection algorithms with the weighted voting mechanism and boost the outlier scores into the training set to form an outlier-adapted training set. To reduce the information loss caused by under-sampling when dealing with imbalanced data, a new bagging-based balanced sampling method is proposed to enhance the traditional under-sampling methods with the bagging strategy to obtain a balanced training set. To further improve the performance of the proposed model, a stacking-based ensemble modeling method is proposed to first perform parametrical optimization and then construct the stacking-based multi-stage ensemble model. Five datasets from the UC Irvine machine learning repository and five evaluation indicators were adopted to evaluate the model performance. The experimental results indicate the superior performance of the proposed model and prove its robustness and effectiveness.