Enhancing Electronic Nose Performance for Differentiating Civet and Non-Civet Roasted Bean Coffee Using Polynomial Feature Extraction Methods

Coffee, a popular beverage worldwide, requires thorough quality assessment to ensure its authenticity and meet consumer demands. Traditional methods in the industry are often subjective, expensive, and time-consuming. This study used a compact, portable electronic nose (e-nose) with machine learning models to classify and distinguish between civet and non-civet roasted beans. The polynomial feature extraction method was used to extract important parameters from the sensor response and improve system performance. Classification models like linear discriminant analysis (LDA), logistic regression (LR), quadratic discriminant analysis (QDA), and support vector machines (SVM) were applied to classify the samples. Among these, the LDA model with polynomial features yielded the highest validation and test accuracies, with values of 0.89 ± 0.04 and 0.93, respectively. This was higher than the statistical feature methods, which obtained validation and test accuracies of 0.80 ± 0.07 and 0.87, respectively. The acquired e-nose results were correlated with compound concentrations in roasted coffee beans measured by gas chromatography–mass spectrometry (GC–MS). These findings demonstrate the e-nose system's promising potential to effectively distinguish civet from non-civet roasted coffee beans based on their aroma profiles using polynomial feature extraction methods.