Anatomy of peer-to-peer (P2P) lending fraud: A review with managerial implications

Peer-to-peer (P2P) lending platforms enable direct matching between borrowers and lenders, expanding access to credit but also amplifying fraud risks, including identity abuse, misrepresentation, collusion, predatory lending, and platform-level misconduct. This paper presents a systematic literature review of 53 studies on fraud detection and prevention in P2P lending, selected using predefined eligibility criteria and screened for relevance to fraud definitions, data and feature practices, modelling approaches, and evaluation designs. Using a thematic and qualitative synthesis of the reviewed literature, we examine how fraud is operationalised, which ML and AI techniques are most effective, and what managerial and ethical implications follow. Across the literature, progress is constrained by scarce and heterogeneous fraud labels, severe class imbalance, limited cross-platform transferability, and temporally evolving fraud behaviour. In response, recent work increasingly combines imbalance-aware learning (cost-sensitive and positive–unlabelled schemes), relational modelling that captures coordinated activity (graph-based features and network representations), and explainability methods to support auditability and user trust. However, evaluation remains uneven: strong within-platform performance often degrades under temporal splits or platform shifts, underscoring the need for drift-aware validation and monitoring. The review qualitatively consolidates a fraud taxonomy for P2P lending and links recurring empirical constraints to deployable model and governance choices, supporting more robust, transparent, and sustainable platform operations in practice today.