In Scientific field, the data collection resulted into large scale growth by continuous additions of data. The characteristics such as velocity, variety, volume, etc. make the collected data as Big data. Analysis of such data uses various data mining techniques. Clustering is one among them and it is an unsupervised learning technique used for statistical data in bioinformatics, social networks, etc. Among various clustering techniques the K -means clustering is a common and predominant algorithm. However, the accuracy of original K-means algorithm heavily depends on centroids at the beginning and it has high computational complexity. In this paper we present an empirical study on enhanced k-means algorithm for high accuracy clustering with the initial centroids selection in an improved manner. © Research India Publications.