A mechanistic review on machine learning-supported detection and analysis of volatile organic compounds for food quality and safety

Food quality and safety have received much more attention in recent years thanks to the increase in food consumption and customer awareness of food quality assurance. Volatile organic compounds (VOCs) detection and analysis techniques are powerful tools for assessing the quality of food products due to their non-destructive, eco-friendly, continuous, and real-time monitoring merits. Machine learning (ML) -supported electronic nose (EN), colorimetric sensor array (CSA), and gas chromatography (GC) hyphened techniques (e.g., GC-MS and GC-IMS) are becoming a hot research area in Food Sciences. In this review, the rationales, advantages, and limitations of these technologies are introduced, as well as ML implementation details in application scenarios. In particular, ML fundamentals of data processing, modeling, and performance evaluation are discussed based on the most recent cases of food VOC detection and analysis studies, followed by the comprehensive applications of ML in different fields of food research including origin traceability, adulteration, quality control, and pathogen detection. With advances in ML, e.g., parallel computing, computer vision, and odor imaging, new food VOC technologies like CSA and EN are replacing traditional GC detection and analysis. Many previously intractable problems in the food industry, e.g., food origin traceability and food adulteration, have been solved by state-of-the-art ML algorithms. However, new challenges in food VOC detection and analysis are emerging, and researchers are exploring new solutions, e.g., edge/cloud computing, EN sensor drifting, and CSA standardized fabrication, to solve more food quality and safety problems.