Machine learning and GC-MS revealing the characteristic flavour compounds of Cheddar cheese origin-independent at different ripening periods.

The complexity of cheese flavour components, different origin and variability in experimental data have hindered credible flavour description of Cheddar cheese at different ripening time periods. This study combined GC-MS with machine learning to explore the common characteristic ingredients of Cheddar cheese independent of origin during ripening stage at 6-12 °C. A random forest model among six classifiers performed best in assessing Cheddar cheese ripening time and 14 flavour substances (ketones, acids, and lactones) were selected as characteristic flavours by recursive feature elimination from 66 flavour substances to train the model. SHAP interaction value and decision trees revealed the least flavour substances and their respective concentrations required to determine ripening time. Finally, commercial Cheddar cheese from various countries in our experiment validated the reliability of characteristic flavour compounds. This study is the first to confirm that certain flavours of Cheddar cheeses from diverse origins share consistent forming mechanisms, advancing research on flavour development and ripening time determination.