Evaluation of beef quality using machine learning based on the CIELAB color space

This study presents a Python-based, machine learning approach to predict spoilage indicators of beef using non-destructive color measurements from the CIELAB color space. A two-stage modeling strategy was employed, where spoilage indicators predicted from a∗ values in the first-stage were used as input features in a second-stage model to estimate spoilage rate. First, a strong correlation between the a∗ value and spoilage indicators, validated through Pearson correlation analysis, highlights the feasibility of color data for quality prediction. Among the regression models tested, the random forest (RF) regression model demonstrated superior performance, achieving R² values of 0.912 and 0.804 for metmyoglobin (Met.Mb) and peroxide value (PV), respectively. Gradient boosting (GB) regression was most effective for pH prediction, while the k-nearest neighbors (KNN) regression model excelled in thiobarbituric acid reactive substances (TBARS) estimation. Then, the RF regression model was further optimized to predict spoilage rates using Met.Mb, pH, PV, and TBARS as input features, achieving an R² of 0.988 and RMSE of 2.120. SHapley Additive exPlanations (SHAP) analysis identified TBARS as the most influential variable, followed by Met.Mb and pH. These findings demonstrate the potential of combining machine learning models with non-destructive methods for real-time beef quality assessment, offering a practical and resource-efficient alternative to traditional analytical approaches.