Interpretable ANFIS Framework for Performance Evaluation and Managerial Rule Extraction in Dynamic Supply Chains

Abstract

Modern supply chains operate in dynamic, multi-stage environments where disruptions in procurement, storage, production, inspection, and assembly propagate across stages, causing cumulative inefficiencies. This paper introduces an interpretable Adaptive Neuro-Fuzzy Inference System (ANFIS) framework designed to enable transparent and adaptive performance governance rather than static optimization. The framework integrates two decoupled fuzzy layers: FIS-1 performs stage-level diagnosis through compact ANFIS models that translate operational signals into tactical rules, while FIS-2 aggregates product-level outcomes into a Strategic Performance Index (SPI) for managerial oversight. The framework is evaluated within an ERP-inspired operational setting that represents a realistic industrial workflow across multiple supply-chain stages. In contrast to conventional neuro-fuzzy approaches, the proposed framework explicitly establishes an interpretable linkage between stage-level performance diagnosis and product-level evaluation within a unified and governance-oriented structure. The results indicate that the framework exhibits adaptive and interpretable behavior in response to operational disturbances, supporting dynamic decision-making across iterative operational cycles. Operationalized within a closed-loop, iterative governance cycle (detect-dispatch-measure-adapt), the framework was evaluated over a 30-day simulated horizon. The rules-enabled dynamic regime demonstrated sustained, cycle-driven performance improvements, including a rolling average 10.3% reduction in total delay, a 1.4 percentage-point containment in cumulative cost deviation, and a 1.85% gain in throughput relative to a static, monitor-only baseline (Section IV). The core methodological contribution is the decoupled, interpretable governance structure that explicitly links stage-level operational diagnostics to product-level strategic decision support within ERP/MES environments. This provides a transparent and auditable pathway for converting AI-generated performance signals into actionable managerial interventions.