Date of Award

4-21-2023

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Information Science

First Advisor

Robert Belford

Abstract

Volatile Organic Compounds (VOCs) are emitted gases that include many chemicals that can be created by biological processes or the byproduct of industrial processes, and can be present in a variety of pollutants. VOCs can cause a variety of ailments in humans and animals, and are expressed in plants as a way of signaling damage or the presence of pests. In this Thesis, an Internet of Things (IOT) E-nose device is outfitted with a gas array of eight Metal Oxide Semiconductor (MOS) type gas sensors, which are sensitive to a broad range of VOCs, yet lack selectivity and are unable to identify chemicals. A Python script was developed to analyze the signals created for three classes of VOCs (alcohols, ketones, cycloalkanes) and isolate these chemical events for machine learning. Three different feature lengths of chemical events were tested with four different supervised classification algorithms. Random Forest classification achieved the highest with 100% accuracy with the shortest feature length. Support Vector Machine and K-Nearest Neighbor achieved 97% and 95% average accuracy across all lengths respectively. Gaussian Naive-Bayes performed the poorest with an average accuracy of 94.8%. These results indicate that a highly accurate classification model can be created to distinguish amongst the three classes of VOCs utilizing these methods.

Download

Share

COinS