Performance comparison of PCA and ICA pre-processors in identification of individual gases using response of a poorly selective solid-state sensor array

Abstract

Identification of odours/gases using an array of chemical sensors is a challenging task. The cross-sensitivity of individual sensors to different types of gases results in poor selectivity. This makes identification of individual gases a computationally demanding task. Unsupervised pattern recognition techniques like principal component analysis (PCA) have widely been used for odour identification. However, a more recent and highly popular pattern analysis technique viz. independent component analysis (ICA) has not been employed widely for odour identification applications. This paper presents performance comparison of PCA and ICA as pre-processing techniques in identification of four different types of gases/odours using published response of a chemical sensor array capable of sensing at room temperature. Raw data pre-processed with PCA and ICA were finally classified using a K-nearest neighbour (KNN) classifier. The efficacy of the techniques has been compared in both the cases. It was observed that performance of ICA as a pre-processing technique has been more consistent. Furthermore, the ICA-KNN classifier was found to give higher success rate and classification efficiency as compared to PCA-KNN classifier at different values of K. © 2012 IFSA.