A Portable Real-Time Electronic Nose for Evaluating Seafood Freshness Using Machine Learning

Real-time evaluation of seafood freshness is essential for ensuring quality and safety. This study presents an electronic nose (e-nose) system designed to assess seafood freshness using gas sensors and machine learning (ML) algorithms. The system detects volatile organic compounds (VOCs) released during spoilage and employs hyperparameter-optimized ML models for both classification (fresh vs. not fresh) and regression (shelf-life prediction). From an initial pool of eight gas sensors, four were selected based on Predictive Power Score (PPS) analysis for their sensitivity to spoilage-related gases: MQ137 (ammonia), MQ2 (alcohol), MQ136 (hydrogen sulfide), and MQ9 (methane). A comprehensive dataset, E-Nose-Based Seafood Freshness (ESF), was developed, comprising 336,997 data points from five major seafood species: tuna, salmon, cod, shrimp, and crab. Various ML models were evaluated, including Random Forest, KNN, Naive Bayes, Decision Tree, AdaBoost, Gradient Boosting, and XGBoost. The XGBoost achieved the best performance with 100% precision, recall, accuracy, and F1-score, and a cross-validation accuracy of 99.8%. It also demonstrated excellent efficiency with detection times under 1 millisecond and memory usage of ~4 kB. For regression, XGBoost maintained strong results ( $\mathrm{R}^{2} = 100$ %, RMSE = 0.08, CV MSE = 0.009) with fast prediction times (~12 milliseconds) and memory usage under 5 kB. These findings confirm the system’s potential as a lightweight, scalable, and accurate solution for real-time seafood freshness monitoring, supporting food safety, quality assurance, and waste reduction in the supply chain.