Machine Learning-Based Rapid Detection of Volatile Organic Compounds in a Graphene Electronic Nose.

Abstract

Rapid detection of volatile organic compounds (VOCs) is growing in importance in many sectors. Noninvasive medical diagnoses may be based upon particular combinations of VOCs in human breath; detecting VOCs emitted from environmental hazards such as fungal growth could prevent illness; and waste could be reduced through monitoring of gases produced during food storage. Electronic noses have been applied to such problems, however, a common limitation is in improving selectivity. Graphene is an adaptable material that can be functionalized with many chemical receptors. Here, we use this versatility to demonstrate selective and rapid detection of multiple VOCs at varying concentrations with graphene-based variable capacitor (varactor) arrays. Each array contains 108 sensors functionalized with 36 chemical receptors for cross-selectivity. Multiplexer data acquisition from 108 sensors is accomplished in tens of seconds. While this rapid measurement reduces the signal magnitude, classification using supervised machine learning (Bootstrap Aggregated Random Forest) shows excellent results of 98% accuracy between 5 analytes (ethanol, hexanal, methyl ethyl ketone, toluene, and octane) at 4 concentrations each. With the addition of 1-octene, an analyte highly similar in structure to octane, an accuracy of 89% is achieved. These results demonstrate the important role of the choice of analysis method, particularly in the presence of noisy data. This is an important step toward fully utilizing graphene-based sensor arrays for rapid gas sensing applications from environmental monitoring to disease detection in human breath.