| Abstract [eng] |
As loan portfolios of individuals in Lithuania continue to grow, the popularity of peer-to-peer lending platforms is growing as well. Both among borrowers and those seeking to borrow. As peer-to-peer lending platforms are also open to investment by retail investors, it is essential that the credit risk of the borrower, which is directly linked to investor losses, is correctly identified. This paper focuses on the assessment of creditworthiness risk on peer-to-peer lending platforms. The first part of the paper analyses the academic literature, discussing methodologies for credit risk assessment, available data sources and commonly used datasets for research through the prism of variables. The second part of the paper presents the research methodology by discussing the selected machine learning algorithms and their operational principles. Five machine learning algorithms were selected on the basis of the literature analysis: logistic regression, decision tree, random forest, XGBoost and LightGBM methods. In the results part of the study, all five classification methods were applied to the loan dataset of a peer-to-peer lending platform operating in Lithuania. The performance of the methods was evaluated using the AUC metric. The results showed that the best method for classifying both solvent and insolvent borrowers was the XGBoost algorithm with an AUC of 0.812. The LightGBM, logistic regression, decision tree and random forest methods performed worse. The results were also used to identify the most significant influencing variables and to compare the results with those obtained in the literature. |
