Classification and Prediction of VOCs Using an IoT-Enabled Electronic Nose System-Based Lab Prototype for Breath Sensing Applications

Abstract

Electronic nose (e-nose) systems are well known in breath analysis because they combine breath printing with advanced and intelligent machine learning (ML) algorithms. This work demonstrates development of an e-nose system comprising gas sensors exposed to six different volatile organic compounds (VOCs). The change in the voltage of the sensors was recorded and analyzed through ML algorithms to achieve selectivity and predict the VOCs. In this work, a novel approach to automatic learning technology that systematically categorizes and implements standard algorithms for use on gas sensors’ data set is presented. Different algorithms were compared based on F1 score, accuracy, and testing time. Performance testing of these methods is also conducted on both a Google Colab and a single-board computer, simulating their application in portable Internet of Things (IoT) sensor systems. Post validation, a simple IoT-enabled prototype was prepared that was tested in the presence of normal breath, alcohol (simulated breath), mint, mouthwash, and cardamom. The model system could classify a simulated breath alcohol sample and other breath samples with an accuracy of 0.96 obtained from the Extra Trees model. This work can be scaled up to a system wherein further breath print analysis can be used for breath diagnostic applications to detect diseases or a person’s physiological condition.