Utilizing drip metabolites and predictive modeling for non-destructive freshness assessment in pork loin

Abstract

This study validated the use of pork drip metabolites for non-destructive freshness prediction. The pork loin was vacuum-packaged and stored for 27 days at 4 °C. The pH, drip loss, total aerobic bacterial counts (TAB), microbial composition and drip metabolites were examined. LASSO and Random Forest (RF) were selected and used for variable selection, while Ridge regression and Support Vector Regression were utilized to develop predictive models. Validation was performed using leave-one-out cross-validation. LASSO and RF selected 13 and 10 metabolites, respectively. The metabolites selected by each method were trained using Ridge regression and SVR. Each of the four trained models achieved R2 values of over 0.9. In the validation step, the model trained by Ridge regression using drip metabolites selected through LASSO showed the lowest RMSE value of 0.283 log CFU/g. Therefore, selected drip metabolites can be used to predict TAB and microbial composition of pork loin through mathematical modeling.