Development of an IoT-Driven Survival Analysis Model Using Supervised Learning for Early Disruption Prediction in Resilient Supply Chains

Modern supply chains face increasing vulnerability to disruptions caused by factors such as supplier delays, equipment failures, logistics bottlenecks, and environmental events. Building resilient supply chains requires not only detecting disruptions after they occur but also predicting them in advance with sufficient lead time for proactive intervention. This study addresses this challenge by developing a novel predictive model that integrates real-time Internet of Things (IoT) data streams with survival analysis and supervised learning techniques for early disruption prediction. A multi-stage methodology is proposed. First, IoT sensor data (e.g., temperature, vibration, location, and operational cycle counts) are collected from critical nodes within a supply chain network. These raw data streams are preprocessed through noise filtering, missing value imputation, and feature extraction to construct time-to-event profiles. Second, a survival analysis framework-specifically a Cox proportional hazards model enhanced with machine learning components-is developed to estimate the conditional probability of disruption occurrence over time. Third, supervised learning algorithms (including Random Survival Forests and gradient-boosted models) are trained on labeled historical data to identify patterns preceding disruptive events. The proposed model is evaluated using a real-world dataset collected from a multi-echelon perishable goods supply chain over a ۱۲-month period. Experimental results demonstrate that the IoT-driven survival model achieves a time-dependent AUC of ۰.۸۹ and a Brier score of ۰.۱۲, significantly outperforming conventional classification-based early warning systems. More importantly, the model provides probabilistic disruption forecasts with an average lead time of ۴.۵ hours before critical thresholds are exceeded, enabling proactive mitigation actions such as rerouting, inventory repositioning, or supplier substitution. The key contribution of this research is the seamless integration of IoT real-time monitoring with survival analysis, transforming passive disruption detection into a proactive, time-aware prediction capability. The proposed framework enhances supply chain resilience by allowing decision-makers to anticipate disruptions before they escalate. Future work will focus on incorporating multi-modal data sources (e.g., social media and weather forecasts) and extending the model to decentralized IoT-edge architectures for real-time inference in large-scale supply networks.

کلیدواژه ها:

Internet of Things (IoT) ، Survival Analysis ، Supervised Learning ، Disruption Prediction ، Resilient Supply Chain ، Early Warning System ، Real-Time Data