From simple classification methods to machine learning for the binary discrimination of beers using electronic nose data

This paper deals with the development and the implementation of an electronic nose based on Metal Oxide Semiconductor (MOS) as an innovative instrument to beer aroma recognition. Principal component analysis (PCA) results revealed a good discrimination, as two clearly separated groups (alcoholic and non-alcoholic brands). Linear discriminant analysis (LDA) was performed on the important variables which were detected based on PCA loadings. The results showed that 100% accuracy for both training and test sets was obtained using those variables. Soft independent modeling of class analogy (SIMCA) and partial least square discriminant analysis (PLS-DA) methods confirmed the PCA and LDA results of classification accuracy and their results were (100%, 100%) and (100%, 100%), for training and test sets, respectively. Finally, support vector machine (SVM) was considered and full accuracy (100%) for beer classification was again achieved for both training and test sets. The results showed that if simple methods perform well they may be preferred.