A decentralized multi-agent framework integrating interpretable machine learning for automated warranty fraud detection in supply chains
Year: 2026
Type: preprint
Abstract: Fraudulent warranty claims pose a substantial challenge in modern supply chain management, leading to significant financial losses, operational inefficiencies and reputational risks for manufacturers. A critical barrier to deploying AI for fraud detection is maintaining human trust in automated decisions, systems must be not only accurate but transparent, explainable and aligned with human oversight requirements. This study proposes a novel human-centric hybrid framework integrating decentralized Multi-Agent Systems (MAS) with interpretable Machine Learning (ML) for automated fraud detection in Warranty Management Systems (WMSs). The framework combines the predictive accuracy of Random Forest (RF), the interpretability of Logistic Regression (LR) and the scalability of MAS, addressing the limitations of traditional rule-based and centralized AI approaches. A controlled proof-of-concept simulation was conducted using a synthetic dataset of 1,000 records designed to emulate real-world warranty claim patterns, where RF achieved 94% accuracy (F1-score = 0.90) and LR achieved 88% accuracy (F1-score = 0.86) in fraud classification. Key fraud indicators identified include claim value, purchase source and product age. The proposed MAS framework demonstrates a 35% reduction in per-claim processing time compared to manual auditing baselines, while supporting scalable, adaptive and autonomous decision-making capabilities. The framework’s decentralized architecture embeds explainability through SHAP analysis and complies with regulatory standards such as GDPR and ISO 27001, ensuring both operational efficiency and ethical accountability. The findings highlight the practical and strategic potential of combining interpretable AI with distributed agent-based systems to enhance operational efficiency and decision support in fraud-prone supply chains. Future research directions include real-world dataset validation with industrial partners, integration with IoT-enabled warranty systems, and benchmarking against deep learning approaches. The framework advances human-centric intelligent systems research by demonstrating how transparency, auditability and regulatory assurance can be operationalized in autonomous industrial AI deployment.
Source: Research Square
Institutions: Al-Aqsa University, Süleyman Demirel Üniversitesi
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7204795385