An Optimized Version of the K-Means Clustering Algorithm

Abstract

Abstract-This paper introduces an optimized version of the standard K-Means algorithm. The optimization refers to the running time and it comes from the observation that after a certain number of iterations, only a small part of the data elements change their cluster, so there is no need to re-distribute all data elements. Therefore the implementation proposed in this paper puts an edge between those data elements which won't change their cluster during the next iteration and those who might change it, reducing significantly the workload in case of very big data sets. The prototype implementation showed up to 70% reduction of the running time.

Accelerated K-means Clustering Algorithm

“…The accuracy of the proposed work is at par with the accuracy of the related works tested in case of Wine dataset. When considering Iris dataset, the accuracy measures obtained are similar for Khan and Ahmad"s work [7], mo re than Enhanced K-means algorithm [11] and slightly less than the P-Kmeans [19]. We could not find accuracy results for ionosphere dataset in any optimized versions of the k-means clustering, so the related portion is mentioned NA (Not available).…”

“…Concepts i and ii have been used by Poteras et al [19] and consequently in this proposed variant of K-means.…”

“…This method suggested by Poteras, Mihaescu and Mocanu [19] uses the concept of distances. Specifically, if the difference of distances of a point fro m its own centroid and other centroids is below a threshold, it is a border point and very likely to change the cluster in next iteration.…”

“…We compared our results with available results of Khan and Ahmad"s work [7], Enhanced k-means [11] and P-Kmeans [19]. Tab les8 and 9 show the tabulated results of the comparison for accuracy and F1 score respectively.…”

Evaluation of an Optimized K-Means Algorithm Based on Real Data

“…

Abstract-In a previous paper [1] we introduced an optimized version of the K-Means Algorithm. Unlike the standard version of the K-Means algorithm that iteratively traverses the entire data set in order to decide to which cluster the data items belong, the proposed optimization relies on the observation that after performing only a few iterations the centroids get very close to their final position causing only a few of the data items to switch their cluster.

…”