An Improved K-means Clustering Algorithm for Classification of Odor/Gas Sensor Data Using Normalized Cosine Distance Parameter

Shivam Choudhary; Ravi Kumar

doi:10.37591/joipprp.v2i2.429

Abstract

This paper presents a novel approach of K-means Clustering for classification of odors/gases (E-nose) using cosine distance as a distance parameter. A sensor array constituting five sensors is exposed to four different types of gases to extract data. The problem of classifying the data into respective classes in considered as a K-means Clustering task. To quantify the amount of similitude between the data corresponding to same classes; usually euclidean distance is used as a distance parameter. The distance parameter used in this paper is normalized cosine distance as the euclidean distance suffers from misclassification when the classes are highly spread in the pattern space. Cosine distance measure the angle as a distance parameter between the data vectors. It was observed that employing cosine distance measure resulted in better or similar classification performance as compared to euclidean distance over ten folds of cross validation scheme. Cite this ArticleShivam Choudhary, Ravi Kumar.An Improved K-means Clustering Algorithm for Classification of Odor/Gas Sensor Data Using Normalized Cosine Distance Parameter. Journal of Image Processing & Pattern Recognition Progress. 2015; 2(2): 56–60p.

Abstract

Keywords