Feedforward Neural Network for Real-time Prediction of Lakatan Banana Shelf Life using Sensor Data from Transport Conditions

Effective monitoring is necessary while transporting perishable items, such as Lakatan bananas, to minimise post-harvest losses. This work introduces a sensor-based machine learning system that uses real-time vibration, temperature, humidity, and ethylene level data to forecast the shelf life of bananas during transit. Due to the limited computing capability of the ESP32 microcontroller, machine learning inference is handled by a backend server. The prototype receives predictions wirelessly and shows the findings on an I2C LCD screen. The system's average prediction accuracy for ripening times was 92.25%. While shorter durations exhibited lower accuracy, predictions for durations greater than 24 hours were especially accurate (90–96%), indicating strong potential for long-haul transport monitoring. Short-term forecasts should be used with caution. The suggested system provides a workable way to reduce spoilage and boost supply chain effectiveness. For wider agricultural usage, future research will concentrate on real-world deployment, increasing sensor integration, honing prediction models, and enhancing user interfaces.