Related Experiment Video
Updated: Aug 6, 2026
Radio Frequency Identification and Motion-sensitive Video Efficiently Automate Recording of Unrewarded Choice Behavior by Bumblebees
Published on: November 15, 2014
From black box to transparency: Explainable artificial intelligence driven origin identification of Xinjiang Nilka
Xiao-Hui Sun1, Lin-Hua Shi2, Xiang-Shuai Li3
1Key Laboratory of Xinjiang Agricultural Product Quality and Safety, Institute of Agricultural Quality Standards and Testing Technology, Xinjiang Academy of Agricultural Sciences, Urumqi 830091, China.
Abstract:
Frequent origin mislabeling of Xinjiang Nilka black bee honey has caused market chaos and harmed the interests of producers and consumers. This study integrated electronic nose (e-nose) technology with explainable artificial intelligence (XAI) to identify the geographical origin of Xinjiang Nilka black bee honey, addressing the "black box" limitation of conventional machine learning models and improving the transparency of honey traceability. The results showed that the ANN model achieved optimal performance, with 82% test accuracy and an AUC value of 0.87. Monte Carlo-based feature selection screened 16 key sensor features to reduce data dimensionality while preserving model performance. Shapley Additive exPlanations and 3D Partial Dependence Plot analyses further interpreted the model's decision mechanism. This study validates the feasibility of combining e-nose and XAI for honey origin identification, providing a novel technical reference for food safety supervision.
