Multimodal Machine Learning for Proactive Supply Chain Risk Management: An AI-Driven Framework Integrating Sensor, Operational, and External Intelligence

Abstract

Global supply chains are becoming increasingly complex and volatile, necessitating high levels of predictive risk management. The study developed a multimodal artificial intelligence model that combines sensor readings, operational indicators, and external intelligence to forecast supply chain risks with unprecedented accuracy. The research utilized complex machine learning algorithms, including logistic regression, random forest, XGBoost, and gradient boosting models, on a dataset of 3,000 supply chain events with 21 variables across various data modalities. The last logistic regression model achieved 100% accuracy in risk classification, which is significantly higher than the set target of 93%. The system utilized 66 engineered features derived from sensor data, operational data, and textual intelligence sources, including social media feeds, news alerts, and system logs. The most critical predictive features were system log message sentiment, social media feed sentiment, and weather condition encoding. The research demonstrates that the multimodal data integration system yields a significant improvement in predictive performance compared with classical single-source methods. The framework can provide a scalable and interpretable real-time risk assessment framework for the supply chain, with substantial implications in Industry 4.0 settings, where proactive risk management is essential for operational resilience.