Food flavor analysis 4.0: A cross-domain application of machine learning

Food flavor analysis 4.0, originating from Industry 4.0, combines machine learning (ML) and food flavor analysis methods. Currently, food flavor analysis mainly depends on sensory evaluation, instrumental analysis, or a combination of both. In recent years, ML has been used effectively in the analysis and prediction of food flavor. However, few research teams have attempted to summarize the research progress in the combination of ML and food flavor analysis. This study focuses on the recent advances in food flavor analysis combined with supervised learning algorithms, including random forest (RF), support vector machine (SVM), k-nearest neighbor (KNN), neural network (NN), deep learning (DL), and hybrid algorithms. The application of ML in the determination of volatile aromatic compounds in meat, fruits, vegetables, and processed and fermented food products maintained a strong prediction stperformance. Both the back propagation neural network (BPNN) and KNN models performed well in the classification, with accuracy values higher than 90%. In contrast, the RF and SVM models delivered satisfactory performance in terms of classification and regression. Notably, the BPNN model achieved the highest classification accuracy in the analysis of extremely complex and similar samples, whereas the SVM model was considered an ideal regression algorithm when measuring a series of meat samples. In summary, food flavor analysis combined with ML has great potential for rapid detection of food additives, quality, and authenticity.