Classification of Sugar Samples From PCA-Reduced Impedimetric Data Using QD Model

This article presents single-model identification algorithm for the classification of sugar sample suitable for designing a portable meter for in-home application. The classification algorithm is based on the impedimetric data of a typical hydrogel taken over a frequency range from 20 Hz to 200 kHz. Principal component analysis (PCA) is used for data reduction and selection of dominating principal components (PCs). Five suitable impedance features are identified to attain maximum accuracy. The work identifies quadratic discriminant (QD) as the most accurate model among 32 different types of machine learning (ML) models for this classification problem. The QD model is trained using 14 904 data points obtained from 414 sugar samples of different concentrations and later tested with 107 sugar samples. It is found that the proposed model can identify sucrose, glucose, galactose, and fructose with true positive rate (TPR) values 100%, 95.8%, 58.3%, and 33.3%, respectively. It clearly indicates the proposed QD-based algorithm and sensing technique is suitable for accurate identification of sucrose and glucose in hexose samples.