Enhanced Default Risk Assessment: The Integration of Outlier Detection, Borrower Network Similarity, and Explainable AI

Abstract

Evaluating credit risk in peer-to-peer (P2P) lending platforms is crucial due to the absence of information typically accessible through conventional banking channels. The financial system’s integrity may be compromised if default risk is not accurately assessed. This research proposes an Outlier Detection with Borrower Network Similarity (ODBN) framework for enhancing the accuracy of credit risk assessment on P2P platforms. To achieve this goal, we suggest incorporating alternative data into conventional credit assessment methodologies. Initially, an unsupervised learning approach is used to differentiate between borrowers who exhibit unconventional behavior and those that follow regular borrowing patterns. The borrower network centrality measures are computed for these two categories of borrowers to provide alternative data. The predictive power of the similarities found in the regular borrower cluster samples is observed to be higher than that of the eccentric borrower cluster. To gain deeper insights into the results, the Shapley values are visualized as a network. The empirical findings on the Lending Club dataset suggest that the ODBN improves the model’s ability to explain and forecast with more precision. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.