Mapping Frozen Fish Quality via Machine Learning for Predictive Spoilage Kinetics Under Subzero Conditions

Frozen storage modulates the progression of key oxidative and nitrogenous reactions within fish muscle. We therefore identify the drivers of quality degradation in filleted whiting (Merlangius merlangus) and Atlantic bonito (Sarda sarda) during 10-month frozen storage at −12, −18, and −24 °C, and to integrate state-of-the-art machine learning architectures to predict deterioration kinetics and shelf-life trajectories. To this end, following blast freezing at −30 °C for 6 h, samples were periodically (0, 2, 4, 6, 8, and 10 months) assessed for biochemical indices—total volatile base nitrogen (TVB-N), trimethylamine nitrogen (TMA-N), thiobarbituric acid (TBA), and free fatty acids (FFA)—in which proximate composition and pH were determined solely on the same day (Day 0). Whiting displayed progressive increases in all indices, yet values at −24 °C remained within regulatory acceptability, supporting a safe storage period of up to nine months. By contrast, Atlantic bonito retained TVB-N and TMA-N values below regulatory thresholds across storage, but TBA exceeded acceptability limits from the second month onward, and FFA rose after month four. Complementing these findings, machine learning (ML) approaches, including Naïve Bayes, Support Vector Machine, Decision Tree, Multilayer Perceptron, and Extreme Gradient Boosting, were implemented to classify species and predict spoilage kinetics, with Extreme Gradient Boosting achieving the highest accuracy (98.9%, κ = 0.978) and Random Forest providing superior regression performance (R2 = 0.986, RMSE = 0.392). ML models consistently identified TVB-N as the dominant predictor for whiting and TBA for Atlantic bonito, correctly capturing the critical time points of 9 months and 2 months, respectively, and highlighting −24 °C as the most reliable condition for preserving quality. These results underscore the potential of ML as a transformative tool for accurate shelf-life prediction and smarter cold-chain management in frozen fish products.