Estimation of Freshness in Seafood based on Machine Learning Algorithms

This work introduces a real-time system for monitoring seafood freshness that integrates a multi-gas sensor hardware module with predictions based on machine learning. A set of MQ-series sensors (NH, HS, CH, CO, Alcohol, H, LPG), combined with temperature and humidity sensing, detects volatile compounds emitted during seafood deterioration. The gathered data is analyzed via a machine learning pipeline where XGBoost, Random Forest, and KNN models execute TVC prediction (regression) and classify freshness into Fresh, Moderate, and Spoiled. XGBoost reached peak performance with a classification accuracy of 99.95% and an R² score of 0.9997 in estimating TVC. The suggested system offers a quick, affordable, and non-invasive method for real-time evaluation of seafood quality, appropriate for markets, cold storage, and supply-chain oversight.