An efficient K-means clustering algorithm for tall data

Abstract

The analysis of continously larger datasets is a task of major importance in a wide variety of scientific fields. Therefore, the development of efficient and parallel algorithms to perform such an analysis is a a crucial topic in unsupervised learning. Cluster analysis algorithms are a key element of exploratory data analysis and, among them, the K-means algorithm stands out as the most popular approach due to its easiness in the implementation, straightforward parallelizability and relatively low computational cost. Unfortunately, the K-means algorithm also has some drawbacks that have been extensively studied, such as its high dependency on the initial conditions, as well as to the fact that it might not scale well on massive datasets. In this article, we propose a recursive and parallel approximation to the K-means algorithm that scales well on the number of instances of the problem, without affecting the quality of the approximation. In order to achieve this, instead of analyzing the entire dataset, we work on small weighted sets of representative points that are distributed in such a way that more importance is given to those regions where it is harder to determine the correct cluster assignment of the original instances. In addition to different theoretical properties, which explain the reasoning behind the algorithm, experimental results indicate that our method outperforms the state-of-the-art in terms of the trade-off between number of distance computations and the quality of the solution obtained.

Analysis of Simple K-Mean and Parallel K-Mean Clustering for Software Products and Organizational Performance Using Education Sector Dataset

“…e input dataset is normalized first and then within the normalized range (0, 1), initial centroids are selected on a random basis. Using min-max similarity measure, the distance is calculated (0, 1) [39]. Ran Vijay and Bhatia [40] have introduced a K-Mean algorithm, items having least frequency.…”

“…ey implement the parallelized Simple K-Mean clustering algorithm for the use of general laboratory. Para Means provide an easy and manageable client server application, written in C# [39].…”

A Cheap Feature Selection Approach for the K-Means Algorithm

“…Among the different advantages, such as the easiness of its implementation, it must be remarked that both phases of Lloyd's algorithm (assignment and update steps) can be easily parallelized [12], which is a major key to meet the scalability of the algorithm [5]. Additionally, there exists a wide variety of speed-ups/approximations to the K-means algorithm, such as different distance prunning approaches [1], [13]- [15], the Minibatch K-means [16], the Boundary Weighted K-means [17] and several coreset techniques [18]- [23].…”

“…One of the main motivations of this work is to improve the scalability of Lloyd's algorithm with respect to the dimensionality of the K-means problem. Moreover, in the literature there are different competitive approximations to the Kmeans algorithm, such as [17], [20], [22], [48], that do not scale well with respect to this factor. The proposed feature selection algorithm will allow the use of these techniques to approximate the solution of the clustering problem on a tractable number of dimensions for them.…”

A Novel Model on Reinforce K-Means Using Location Division Model and Outlier of Initial Value for Lowering Data Cost

“…Nevertheless, K-means is completely dependent on the centroid of the initial clustering, whose selection causes widely differences in the execution time of clustering–repetition and the clustering results. In the K-means investigated in existing studies, the user arbitrarily determines the number of clusters to categorize initial data, and it leads to classification costs [ 26 , 27 , 28 , 29 ]. As data become more and more diverse in form and bigger and bigger in size, the fast execution that is the advantage of K-means cannot be maximized in the era of big-data [ 30 , 31 , 32 , 33 , 34 ].…”