Classification of Urine Odour Using Machine Learning Methods

Abstract

This paper presents an odour sensing device with machine learning algorithms that can classify urine odour to aid incontinent individuals. The device contains custom made metal oxide sensors that are controlled by a Teensy 3.6 microcontroller. The gas classification experiment was performed with an automatic test rig on four compounds, acetone, ammonia, ethyl acetate and synthetic urine; at five concentration levels and three humidity levels. The collected data were processed employing three classifier methods, k-nearest neighbour (KNN), a shallow neural network (MLP) and a convolutional neural network (CNN). The overall classification accuracies of these three models are 93.5%, 92.6% and 95.4%, respectively. More importantly, both KNN and CNN have 100% success rate in urine classification, and only one misclassification of synthetic urine occurred with the shallow neural network.