Abstract
In this study, machine learning on a microcontroller-based electronic nose is proposed to identify danger in gas mixtures of nitrogen dioxide, carbon monoxide, and formaldehyde. This electronic nose is low-powered and low-cost, making new areas for pollution detection possible. This study researches the use of machine learning in combination with low-cost chemical sensors. The machine learning can filter the noise in the sensors and to some extent compensate for the influence of disturbances such as a change in humidity, temperature, and cross-sensitivity. With reliable, cheap sensors it is possible to produce low-cost electronic noses that may detect environments which represents a health risk. By using a classification approach, it is possible to identify the compound of a threat and separate safe environments from dangerous. A Feed Forward Neural Network (FFNN) and a Random Forest Classifier (RFC) demonstrates a 96 % accuracy to identify harmful concentrations of nitrogen dioxide, carbon monoxide, and formaldehyde using the datasets collected during this study.