This study presents a systematic literature review evaluating the current state of artificial intelligence (AI) applications in combating food fraud. Following the Kitchenham framework, 69 primary studies from peer-reviewed journals across four academic databases were identified and analyzed. The aim was to examine the types of fraud detected by AI, the food products involved, the specific AI techniques used, and the performance evaluation metrics utilized. Most of the included studies focused on the detection of adulteration and mislabeling, particularly origin and quality mislabeling, with spices, herbs, meat and dairy being the most frequently investigated food product categories. Machine Learning (ML) and Deep Learning (DL) were the primary approaches utilized, ML was the most dominant, with Support Vector Machines (SVMs) and Convolutional Neural Networks (CNNs) emerging as the most frequently applied algorithms. Regarding the data types, spectral and imaging data were predominantly used, and most models were developed using supervised approaches. Nevertheless, despite strong model performance with data gathered from controlled environments such as labs, issues such as data availability and interpretability remain. The findings underscore the importance of AI applied in food fraud detection and the need to explore underrepresented fraud types and food categories.
Fragkos et al. (Wed,) studied this question.