Adversarial Training for Tabular Credit Scoring: A Multi-Attack Robustness Evaluation in P2P Lending

RESUMO

Machine learning-based credit scoring is increasingly central to Peer-to-Peer (P2P) lending, yet its resilience to adversarial manipulation remains poorly understood. We address this with a systematic train-test robustness benchmark on a large Lending Club subset, spanning three model families and four attacks confined to applicant-mutable features. Adversarial training sharply improves robustness against the attack it is trained on and transfers well within the gradient-based family, but transfers weakly to non-gradient corruption, so single-attack defences overstate real-world resilience. Mixed training delivers the most balanced robustness while preserving clean-test performance.