Privacy-Preserving Chargeback Intelligence for Tokenized Payment Systems

Abstract

Chargeback handling in modern payment systems remains heavily dependent on post-transaction workflows that expose more payment data than necessary and often rely on PAN-linked retrieval paths. This paper presents a privacy-aware post-transaction architecture for tokenized payment sys- tems that combines chargeback intelligence, selective evidence disclosure, and audit-oriented work- flow design. Beyond system integration, the paper introduces a formal role-based selective-disclosure policy model, a token-lineage risk formulation for dispute scoring, and an evidence-quality framework that separates completeness, relevance, and verifiability. The proposed framework contains a token- aware chargeback intelligence layer, a privacy-preserving evidence layer, and a compliance-aware audit layer to support PAN-less dispute handling wherever feasible. Public transaction datasets are used as the empirical basis, with token-related post-transaction structures constructed through controlled augmentation because the source datasets do not natively contain tokenized dispute workflows. The evaluation uses PaySim as the base transaction backbone, the ULB/Worldline credit-card fraud benchmark for fraud-context support, and a chargeback-labeled dataset for dis- pute modeling. In the primary chargeback prediction experiment, the token-aware model achieved an AUC of 0.9317 and an F1-score of 0.6710, outperforming both a static rule baseline and a ba- sic machine-learning baseline. Additional analysis includes ablation testing, sensitivity analysis, repeated-trial workflow evaluation, and failure-oriented threat scenarios. The proposed workflow reduced the weighted privacy-exposure score by 79.1% relative to a PAN-oriented baseline while improving evidence quality at the cost of higher operational overhead. These findings suggest that token-aware post-transaction intelligence can improve dispute prediction and reduce sensitive-data exposure simultaneously. The paper contributes a practical architecture, formalized workflow logic, explicit metric definitions, and a stronger evaluation framework for privacy-preserving chargeback operations in tokenized payment systems.