Machine learning and flavoromics-based research strategies for determining the characteristic flavor of food: A review

Background

Flavor is an important indicator of food quality. In recent years, machine learning (ML) has been widely used in food feature flavor mining and analysis. However, case summaries and ML, and flavoromics analyses are lacking.

Scope and approach

This paper highlights recent advances in the joint application of food flavoromics analysis and ML algorithms, including detection techniques commonly used in food flavor research and data analysis methods for different ML models. These techniques are analyzed and compared, and their advantages and limitations are discussed.

Key findings and conclusions

The application of ML in the detection of food flavor compounds can produce strong analytical and predictive performance. Each ML model has its own advantages and disadvantages. Models such as k-nearest neighbor (KNN), support vector machine (SVM), decision tree (DT) and deep learning (DL) can handle complex and larger datasets but have high data volume requirements, require time-consuming training, and are prone to overfitting. Models such as principal component analysis (PCA), partial least squares (PLS), and random forest (RF) are relatively simple, have a low data volume requirement, and can be trained quickly but may suffer from underfitting when dealing with complex data. When multiple ML models are used together to predict flavor, the model with the highest accuracy or that is better suited for the prediction task can be quickly identified. In conclusion, the combination of food flavor analysis and ML has great potential for specialty food flavor mining, quality assessment, and authenticity.