<![CDATA[Classifying tea aromas using an electronic nose (e-nose) system offers rapid, non-destructive quality assessment. However, metal oxide semiconductor (MOS)-based gas sensors are often affected by temperature and humidity, reducing classifier robustness. To address this, we propose an environmental compensation approach to boost robustness in machine learning-based tea aroma classification. Specifically, we analyzed 400 tea samples (100 per class: black, green, red, yellow) using an e-nose with 10 MOS sensors under three scenarios: (A) sensor features only, (B) integration of temperature–humidity features, and (C) temperature–humidity-based signal compensation before feature extraction. For classification, we used SVM, Random Forest, KNN, and a soft voting ensemble. Notably, Scenario C performed best, achieving 85.00% accuracy with SVM. Furthermore, robustness analysis revealed that KNN led on test data (RI=0.9851), while SVM was perfect in cross-validation (RI=1.000). These results confirm that environmental compensation effectively improves the MOS e-nose system stability.]]>
Copyrights © 2026
