Application of machine learning and deep learning to detect adulteration in food flour based on spectroscopy data: A systematic review

Food quality and safety are very important aspects in the food industry, but counterfeiting often occurs, especially in powdered products. The development of non-destructive technology based on spectroscopy, combined with machine learning and deep learning algorithms, is increasingly being applied to quickly and accurately detect adulterants. This study aims to identify, analyze, and review research trends related to the detection of adulterated powdered food products by combining spectroscopy technology and machine learning or deep learning methods through a systematic literature review (SLR) approach. The study identified 32 out of 105 articles selected from Scopus, Web of Science, and PubMed. The research trend shows a significant increase since 2019, dominated by the Asian region. Commonly used spectroscopy technologies include NIR, Vis-NIR, and Raman, combined with algorithms such as PLSR, CNN, and SVM to improve detection accuracy beyond 90% and achieve high R-squared (R²) values. Data pre-processing techniques, such as filtering, have also proven effective in improving analysis results. The development of intelligent detection systems using machine learning models, data augmentation techniques, and transfer learning, along with multidisciplinary collaboration between food science, computer science, and instrumentation fields, will strengthen future research.