Development of machine learning-based shelf-life prediction models for multiple marine fish species and construction of a real-time prediction platform

At least 10 million tons of seafood products are spoiled or damaged during transportation or storage every year worldwide. Monitoring the freshness of seafood in real time has become especially important. In this study, four machine learning algorithms were used for the first time to develop a multi-objective model that can simultaneously predict the shelf-life of five marine fish species at multiple storage temperatures using 14 features such as species, temperature, total viable count, K-value, total volatile basic-nitrogen, sensory and E-nose-GC-Ms/Ms. as inputs. Among them, the radial basis function model performed the best, and the absolute errors of all test samples were <0.5. With the optimal model as the base layer, a real-time prediction platform was developed to meet the needs of practical applications. This study successfully realized multi-objective real-time prediction with accurate prediction results, providing scientific basis and technical support for food safety and quality.