Machine learning for food flavor prediction and regulation: models, data integration, and future perspectives

Area of Science:

• Flavor science and food quality analysis.
• Application of machine learning (ML) in food research.
• Integration of analytical chemistry and data science.

Background:

• Traditional flavor evaluation methods are subjective and limited in scalability.
• High-throughput technologies and multimodal datasets necessitate advanced analytical approaches.
• Machine learning (ML) offers a powerful solution for complex flavor system analysis.

Purpose of the Study:

• To review current flavor detection techniques and ML applications in diverse domains.
• To compare various ML models including supervised learning, ensemble algorithms, and deep learning.
• To discuss the role of data dimensions, future prospects, and explainable AI in flavor science.

Main Methods:

• Review of supervised learning (SVM, DT), ensemble (XGBoost, LightGBM), and deep learning (CNN, ANN) models.
• Analysis of integrated systems combining electronic nose (E-nose), electronic tongue (E-tongue), GC-MS, and GC-IMS.
• Incorporation of explainable AI (XAI) tools like SHAP for model transparency.

Main Results:

• ML enables precise flavor prediction, compound screening, and real-time process control.
• Ensemble and deep learning models excel with complex, nonlinear datasets.
• XAI tools enhance model transparency by linking predictions to features, improving accuracy and generalizability.

Conclusions:

• ML opens new technological avenues in flavor science for predicting and controlling flavor formation.
• ML aids in verifying product authenticity and designing targeted flavor compounds.
• Innovations like attention mechanisms and digital twins support dynamic flavor modulation and genotype-phenotype relationship identification.