K-Means Algorithm: An Unsupervised Clustering Approach Using Various Similarity/Dissimilarity Measures

Clustering is an unsupervised method of classifying data objects into similar groups based on some features or properties usually known as similarity or dissimilarity measures. K-Means is one of the most popular clustering methods that come under the hard clustering group. In this clustering method, any data object can belong to a single cluster. On the other hand, in soft clustering methods (e.g., fuzzy c-means clustering), the data object can be clustered in more than one cluster with some degree which is specified by the membership value with the limitation imposed as the summation of these membership values should be equal to individual. While the clustering method of K-Means is a comparatively old technique, it still has tremendous popularity in terms of being used in applications for data grouping and machine learning. In this article, K-Means approach with five different distance measures such as Euclidean, Squared Euclidean, Half Squared Euclidean, Cosine, and City Block distance has been explored. A comparative study is made based on the performance of these similarity criteria on real-time Edible oil dataset acquired using MIR spectroscopy. In addition, it attempts to investigate the measure of similarity for a specific collection of unique patterns carrying data. In the MATLAB R2015b environment given by Mathworks, the K-Means algorithm with different similarity and dissimilarity measures were formulated and implemented.