EnDairy: An Ensemble Learning-based Framework for Efficient Dairy Product Identification

Food safety has been a worldwide concern, and raw materials’ adulteration is oneof the significant factors contributing to this problem. Camel milk has high nutri-tional and economic value but faces adulteration problems. With time-consuming,poor stability, and low accuracy, the existing adulteration identification frame-works are limited applications. This paper proposes an efficient dairy adulterationidentification framework (EnDairy) based on an ensemble learning algorithmto solve these problems. Specifically, samples are identified by comparing theprobabilistic prediction values obtained and the optimal threshold by data pre-processing and weak learners iteration fusion into a strong learner. In this paper,we used a batch of camel milk data to verify the effectiveness of EnDairy in realscenarios. We tested its discriminative ability for data with missing values byfeature mask. The results show that EnDairy performs well; the AUC and Recallvalues are equal and 0.986 and 0.981 for the 10% and 20% feature mask treat-ments, respectively. The time taken by the identification model is only 0.637sand 0.648s. EnDairy ensures the accuracy of identification and effectively reducesthe economic and time costs, which can provide supportive information for camelmilk regulation, promote dairy product safety, and advance health for all.