Detection and Accurate Classification of Mixed Gases Using Machine Learning with Impedance Data

Abstract

An inexpensive and effective technique based on machine learning (ML) algorithms with impedance characterization to sense and classify mixed gases is presented. Specifically, this method demonstrates that ML algorithms can distinguish hidden and valuable feature information such as different gas molecules from surface-charged activated carbon fibers. The feature information used for ML is obtained by measuring the impedance and fitting the measured values to an equivalent circuit model. The mixed gases are classified using such feature information to train various automatic classifiers. The collected data consist of the resistances and capacitances extracted from best fitting results in Cole-Cole plots, and they are 5D vectors. The data processed with unsupervised learning are clustered, evaluated with Silhouette scores, and then the unique hidden patterns of individual gases in the mixed gases are obtained. When the supervised ML algorithm, k-nearest neighbor classifier, is used for the analytical features, all combinations of gases have 94% classification accuracy, demonstrating the superiority of the proposed technique.