Machine Learning Approaches for early Prediction of Supply-Chain Disruptions and Operational Risks

Abstract

The frequency and effect of supply-chain disruptions have increased in intensity and impact because of the growing global interconnectivity and complexity of their operation. Although the classic frameworks of supply chain risk management focus on the reactive resilience strategies, the innovations in machine learning provide the prospects of proactive early disruption identification. This paper introduces a systematic conceptual overview of machine learning solutions to early detection of supply-chain disruption and operational risks. The current approaches are divided into supervised learning models, time-series deep learning methods, graph-based network approaches, and anomaly detection systems. The review also examines the data usage trends and metrics of evaluation performance, and specifically those of early-warning performance and lead-time performance. The results have shown that despite the enhancements in the predictive accuracy, there has been little focus on multi-tier network modeling, integration of multimodal data and standardized early-warning evaluation models. This paper is based on the premises of the Dynamic Capabilities Theory, and the conceptual theory of predictive analytics as a strategic facilitator of proactive and adaptive supply chain risk management.