Machine Learning–Based Spoilage Risk Prediction System for Cold Chain Monitoring Using Sensor Data and SHI Modelling

Cold chain monitoring systems are essential for preserving the quality of temperature-sensitive products, yet most existing solutions rely on reactive threshold-based alarms that fail to anticipate spoilage. This paper presents a predictive spoilage risk assessment framework using environmental sensor data and a domain-driven Spoilage Health Index (SHI). Logistic Regression and Random Forest classifiers are comparatively evaluated using only raw temperature and humidity measurements to avoid data leakage and ensure fair model assessment. Experimental results show that while Logistic Regression provides base line interpretability, Random Forest achieves superior predictive performance with accuracy above 92% and ROC–AUC exceeding 0.92, while minimizing false negatives. The SHI is additionally used as an interpretability layer to visualize cumulative environmental stress over time. The proposed approach enables proactive and explainable spoilage risk prediction suitable for real-world cold chain deployment.