Shelf-Life Prediction of Shrimp Gravlax Using Machine Learning: Integrating Traditional Processing with AI Modeling

Abstract

This study aimed to develop shrimp gravlax (Penaeus japonicus) as a ready-to-eat seafood product and to determine its shelf life. The product was prepared using a curing method and stored at 4 degrees C for 30 days. Quality changes were monitored at five-day intervals through analyses of TVB-N, TBARs, peroxide value, pH, water activity, total mesophilic aerobic bacteria, and total psychrophilic bacteria. Gradual shifts in quality parameters were observed during storage, with notable increases in TVB-N, lipid oxidation markers, and microbial counts. Sensory scores declined over time, yet the product remained acceptable until approximately day 25. These findings suggest that shrimp gravlax has a shelf life of around 25 days under the studied conditions. To support freshness evaluation, machine learning models including Support Vector Machine (SVM), K-Nearest Neighbors (K-NN), and Decision Tree (DT) were applied. After data augmentation and parameter optimization, the models achieved high classification performance, reaching up to 100% under optimized conditions. The classification outcomes aligned well with experimental observations, highlighting the potential of machine learning to strengthen shelf-life assessment when multiple quality indicators are considered together. Nevertheless, the models were developed under a single storage condition and focused on classification rather than time-series prediction. Further research using independent datasets and varied storage environments will be necessary to enhance model generalizability. In conclusion, shrimp gravlax can be regarded as a promising ready-to-eat product. Combining traditional processing methods with machine learning provides a practical and innovative approach to shelf-life evaluation in seafood systems.