Explainable AI for Fraud Detection in Decentralized Loyalty Platforms

Blockchain-based loyalty platforms can enhance brand loyalty and boost customer retention, yet like any software they remain vulnerable to fraud. The paper by Teoman Berkay Ayaz, Muhammet Furkan Özara and Ahmet Erkan Çelik (Next4biz R&D Center) and Akhan Akbulut (Next4biz R&D Center and Istanbul Kültür University) presents a transparent fraud-detection framework for PointXchange. PointXchange is a blockchain-based customer loyalty platform that lets brands collaboratively manage and exchange reward points in a decentralized environment; we described the idea behind it in our post on customer loyalty with blockchain and smart contracts.

The study evaluates three gradient boosting machines and two neural networks on one public dataset (Kaggle Ethereum Fraud Detection) and one synthetic dataset generated on the PointXchange prototype. To interpret the model's decisions it applies SHAP, LIME and LRP, and in addition introduces a surrogate-based fidelity assessment.

The top-performing predictors reached F1-scores of up to 0.89 and 0.93 on the two datasets, while the top-performing explanations yielded a fidelity score of complete agreement. The results demonstrate that explainable fraud detection can contribute to both security and transparency in blockchain-based loyalty systems; the study does not stop at empirically evaluating existing approaches but tailors them to a real-world system.

The paper was presented at ASYU 2025 (Innovations in Intelligent Systems and Applications Conference), held in Bursa on September 10-12, 2025, where it received the third best paper award, and appears in the conference proceedings published by IEEE (pp. 1-6).