Gases/odors classification using K-means, hierarchical clustering and self organizing map

Abstract

This work presents the results of the experiments performed to classify the gases/odors with K-means clustering, hierarchical clustering and self organizing map using thick film gas sensor array responses. Already published responses of thick film sensor array for different gases/odors viz. H2, LPG, acetone and 2-propanol are used. Principal Component Analysis (PCA) was used for preprocessing the raw data in order to seek the improvement in classification accuracy. The PCA preprocessed data showed improved cluster visualization with classification accuracy of 97.9%, 97.9%, 100% and 100% as compared to raw data (88.5%, 88.5%, 94.8%, 90.2% with SOM, K-means clustering and hierarchical clustering and supervised-SOM respectively. Also, Euclidean distance parameter with random initialization showed better results than Manhattan distance for the all experiments. It can be concluded that SOM, K-means clustering and hierarchical clustering can be used for unsupervised cluster visualization and classification measures for gas sensing applications. The proposed methods have potential to be utilized in electronic nose (e-nose) applications. © 2015 Taylor & Francis Group, London.