Automated Tuna Freshness Assessment via Gas Sensors and Machine Learning Algorithms

Ensuring the safety and quality of fish products is essential for public health, especially in countries like Indonesia, where tuna is a popular fishery product. Tuna freshness is a critical indicator of seafood safety, directly influencing both its nutritional value and the risk of contamination. This study explores the application of machine learning models to accurately and efficiently classify tuna freshness based on gas emissions detected by an electronic nose (e-nose) system. The system employs MQ-135, MQ-9, and MQ-2 gas sensors, which are sensitive to spoilage-associated gases such as ammonia, methane, and alcohol. A dataset of 58,389 records was generated to capture the gas profiles released by tuna over time, creating a robust dataset for model training and validation. Three machine learning algorithms―K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Naive Bayes―were compared to evaluate classification accuracy and practical applicability. Results indicate that both KNN and SVM achieved high accuracy rates of 99%, while Naive Bayes reached 90%, underscoring the suitability of KNN and SVM for tuna freshness detection. These findings demonstrate the potential of integrating e-nose technology and machine learning to provide a rapid, reliable, and scalable solution for freshness assessment in the fishing industry, enhancing quality control from suppliers to retailers.