Image-Based Machine Learning for Predicting Acceptability Limits in Frozen Pizza Shelf Life

Shelf life of frozen foods is intrinsically linked to consumer sensory acceptability. However, quantifying the synergistic impact of extended storage and variable thermal cycles on perception remains challenging. This study proposes a non-destructive image-based approach for estimating the acceptability of frozen pizza using a machine learning model and identifying tomato sauce degradation as indicator of product quality decay. Qualitative consumer feedback (90%) identified tomato sauce saturation as the primary driver of visual rejection. Image processing pipeline was developed to isolate the sauce region from each sample for further color extraction (saturation in the HSV color space). A second-degree polynomial regression model was used to describe the saturation trend over time and, in parallel, a logistic regression classifier was trained to predict binary consumer acceptability based on both saturation and storage duration. The models were evaluated using frozen pizzas (-12 and -18 °C) for up to 200 days. The regression model achieved an R² of 0.68 and an RMSE of 12.8, while the classifier attained an accuracy of 88.2% and an AUC of 0.93. The resulting framework enables early, non-invasive estimation of product acceptability and shows strong potential for practical application in shelf life studies within the frozen food industry.