Development and Evaluation of a Fraud Detection Method Using Likelihood Ratio and Correlation in Food Traceability

In recent years, food adulteration—the mixing of one brand of meat with another brand of meat—is a significant issue. One approach to detecting such fraud is to monitor the relationship between livestock weight and the weight of its edible parts through a food traceability system. However, conventional methods often fail when fraudsters carefully adjust the weight transferred from one brand to another so that both brands remain within normal ranges. In this study, we propose two fraud detection methods that focus on distributional changes in both livestock weight and edible-part weight. Specifically, one uses a likelihood ratio to detect shifts caused by continuous adulteration, and the other uses correlation to detect a change from very weak correlation to a strong negative correlation caused by brand-to-brand meat transfers. Evaluation experiments show that when the frequency of adulteration exceeded 17.2%, our method achieved the target true positive rate. These results show the effectiveness of our proposed approach.