Abstract—Credit risk estimation and the risk evaluation of credit portfolios are crucial to financial institutions which provide loans to businesses and individuals. Non-performing loan (NPL) is a loan type in which the customer has a delinquency; because they have not made the scheduled payments for a time period. NPL prediction has been widely studied in both finance and data science. In addition, most banks and financial institutions are empowering their business models with the advancements of machine learning algorithms and analytical big data technologies. In this paper, we studied on several machine learning algorithms to solve this problem and we propose a comparative study of some of the mostly used non performing loan models on a customer portfolio dataset in a private bank in Turkey. We also deal with a class imbalance problem using class weights. A dataset, composed by 181.276 samples, has been used to perform the analysis considering different performance metrics (i.e. Precision, Recall, F1 Score, Imbalance Accuracy (IAM), Specificity). In addition to these, we evaluated the performance of the algorithms and compared the obtained results. Also, we studied on explainability of the benchmarked techniques with several eXplainable Artificial Intelligence tools. According to these performance metrics, LightGBM gave the best results among the logistic regression, support vector machines, random forest classifier, bagging classifier, XGBoost and LSTM for the dataset.
Index Terms—Non performing loans, non performing loan prediction, big data, machine learning, supervised learning, explainable artificial intelligence.
The authors are with Yapi Kredi Technology, Turkey (e-mail: salih.imece@ykteknoloji.com.tr).
Cite: Sefik Ilkin Serengil, Salih Imece, Ugur Gurkan Tosun, Ege Berk Buyukbas, and Bilge Koroglu, "A Comparative Study of Machine Learning Approaches for Non Performing Loan Prediction with Explainability," International Journal of Machine Learning and Computing vol. 12, no. 5, pp. 208-214, 2022.
Copyright @ 2022 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).