Abstract

Food waste from early or undetected spoiling is still a major worldwide issue that has a big impact on environmental sustainability, economic efficiency, and public health. Static expiry labels frequently result in the needless disposal of food or the consumption of contaminated food because they do not take dynamic storage conditions into account. We present an intelligent, real-time food expiry monitoring system that combines edge-compatible machine learning with inexpensive IoT sensing to overcome this constraint. The prototype captures spoilage indicators from common perishables like dairy and poultry by integrating temperature, relative humidity, and gas-sensitive modules that target ammonia and hydrogen sulphide. In controlled trials, a lightweight supervised model that was trained on time-series sensor data correlated with microbial growth benchmarks predicts remaining shelf life with 93.8% accuracy. In controlled trials, a lightweight supervised model that was trained on time-series sensor data correlated with microbial growth benchmarks predicts remaining shelf life with 93.8% accuracy. Our approach ensures user privacy, low latency, and offline functionality by performing inference locally on an ESP32 microcontroller, in contrast to cloud-dependent architectures. The system allows for proactive household or small-scale commercial management of food freshness by communicating alerts through a lightweight mobile interface. This study shows that condition-based expiry estimation, as opposed to calendar-based labelling, can greatly reduce preventable waste while improving consumer safety—providing a scalable, reasonably priced, and comprehensible substitute for current methods.