BME688 eNose system for classifying olive oils by quality

Abstract

This work presents the development and evaluation of an electronic nose based on BME688 sensors for the classification of virgin and extra virgin olive oils. The system combines a modular measurement chamber, programmable thermal profiles, temporal feature extraction, and machine learning techniques. From the responses of three sensors and ten thermal steps per sensor, 180 features were extracted for each sample. Linear and non-linear classifiers were evaluated using stratified cross-validation, together with different feature selection strategies. The best performance was obtained with a linear SVM combined with SelectKBest and filtering of highly correlated variables, reducing the input space to 26 features and achieving a balanced accuracy of $0.822 \pm 0.058$. The results show the potential of the proposed system to discriminate between both categories, although validation across independent repetitions highlights the need to improve robustness against experimental variability and sensor drift.