A Machine Learning Approach to Supply Chain Vulnerability Early Warning System: Evidence from U.S. Semiconductor Industry

Abstract

This paper presents a machine learning-based early warning system for detecting and predicting defects in semiconductor devices. This study integrates network research models with advanced machine learning to develop a comprehensive framework for supply chain risk assessment and mitigation. The system can be integrated with multiple data streams, including real-time measurement data, performance measurement equipment, and business indicators, achieved through a combination of combined with Graph Neural Networks (GNN) and Long Short-Term Memory (LSTM) networks. The system achieved 94.3% accuracy in predicting product impact, with an average time of 15.3 days for major events. The research methodology included widespread use across 158 semiconductor manufacturers over 18 months, demonstrating a 64% reduction in impact over time and generating cost estimates of $37.2 million. The hybrid model architecture, combining GNN with LSTM networks, outperformed traditional methods with a precision rate of 0.948 and a return of 0.951. This study contributes to the understanding of supply chain vulnerabilities through the innovative use of network research and machine learning, while developing operational strategies for real-time risk assessment and reductions in semiconductor supply chains.