Although many chemical gas sensors report high sensitivity towards volatile organic compounds (VOCs), finding selective gas sensing technologies that can classify different VOCs is an ongoing and highly important challenge. By exploiting the synergy between virtual electronic noses and machine learning techniques, we demonstrate the possibility of efficiently discriminating, classifying, and quantifying short-chain oxygenated VOCs in the parts-per-billion concentration range. Several experimental results show a reproducible correlation between the predicted and measured values. A 10-fold cross-validated quadratic support vector machine classifier reports a validation accuracy of 91% for the different gases and concentrations studied. Additionally, a 10-fold cross-validated partial least square regression quantifier can predict their concentrations with coefficients of determination, R2, up to 0.99. Our methodology and analysis provide an alternative approach to overcoming the issue of gas sensors' selectivity, and have the potential to be applied across various areas of science and engineering where it is important to measure gases with high accuracy.
Keywords: gas sensors; indoor air quality; machine learning; quantification; selectivity; virtual arrays; volatile organic compounds.