Classification of Animal and Plant-Based Edible Oils with Electronic Nose Using Extreme Gradient-Boosted Decision Tree

Over the recent timeframe, the food industry has seen a growing demand for accurate and efficient methods to classify edible oils, particularly in distinguishing between animal and plant-based oils to ensure authenticity and combat false labelling. This study aims to implement an electronic nose (e-nose) system integrated with an Extreme Gradient-Boosted Decision Tree (XGBoost) Classifier to classify edible oils as either animal or plant-based. The system utilizes an array of metal-oxide semiconductor (MOS) gas sensors to detect volatile organic compounds (VOCs) emitted by the oils, with data preprocessing involving Principal Component Analysis (PCA) for dimensionality reduction. A total of 100 samples, comprising coconut oil, canola oil, pork lard, and beef tallow, were used for training and testing the model and a total of 24 new and unseen samples were used for model validation. The e-nose system achieved an overall accuracy of $\mathbf{9 5. 8 3 \%}$ in classifying edible oils, as validated by a confusion matrix. These findings highlight the potential of the e-nose system combined with XGBoost for applications in food safety, halal authentication, and consumer protection, offering a resource-efficient alternative to traditional methods. Future studies may explore multiclass classification of specific oil types and further enhancements to the e-nose system design.