Using machine learning method to predict the quality of fermented krill sauce based on the freshness of Antarctic krill.

Abstract

Antarctic krill, a shrimp-like crustacean, whose quality strongly influences the quality of fermented shrimp sauce, yet the relationship and prediction methods remain unclear. This study evaluated 75 krill samples of varying freshness and their corresponding shrimp sauces, and successfully developed a high-performance predictive model linking raw material quality to product quality. Shrimp spoiled significantly after only 24 h at room temperature, exhibiting darkened color, increased spoilage-related compounds, with yield and conversion rate sharply decreasing from 80.4 % and 87.1 % to 58.9 % and 63.7 %, respectively. Machine learning with multiple feature selection methods and models showed that Lasso regression effectively reduced collinearity, and its integration with a linear model enhanced prediction accuracy, with nearly 80 % of features well-fitted and about 70 % showing mean absolute percentage error below 5 %. This study successfully established an accurate model linking krill freshness to shrimp sauce quality, which offered guidance for krill utilization and intelligent fermentation.