New approach for clustering of big data: DisK-means

The exponential growth in the amount of data gathered from various sources has resulted in the need for more efficient algorithms to quickly analyze large datasets. Clustering techniques, like K-Means are useful in analyzing data in a parallel fashion. K-Means largely depends upon a proper initialization to produce optimal results. K-means++ initialization algorithm provides solution based on providing an initial set of centres to the K-Means algorithm. However, its inherent sequential nature makes it suffer from various limitations when applied to large datasets. For instance, it makes k iterations to find k centres. In this paper, we present an algorithm that attempts to overcome the drawbacks of previous algorithms. Our work provides a method to select a good initial seeding in less time, facilitating fast and accurate cluster analysis over large datasets.