Feature Ranking of Large, Robust, and Weighted Clustering Result

Saarela, Mirka; Hämäläinen, Joonas; Kärkkäinen, Tommi

doi:10.1007/978-3-319-57454-7_8

Cited by 9 publications

(10 citation statements)

References 17 publications

(25 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The SSE values were computed with formula (1) for the whole data. Finally, statistical comparison between the methods was performed with the nonparametric Kruskal-Wallis test [48,49], since in most cases the clustering errors were not normally distributed. The significance level was set to 0.05.…”

Section: Methodsmentioning

confidence: 99%

Improving Scalable K-Means++

2020

Self Cite

View full text Add to dashboard Cite

Two new initialization methods for K-means clustering are proposed. Both proposals are based on applying a divide-and-conquer approach for the K-means‖ type of an initialization strategy. The second proposal also uses multiple lower-dimensional subspaces produced by the random projection method for the initialization. The proposed methods are scalable and can be run in parallel, which make them suitable for initializing large-scale problems. In the experiments, comparison of the proposed methods to the K-means++ and K-means‖ methods is conducted using an extensive set of reference and synthetic large-scale datasets. Concerning the latter, a novel high-dimensional clustering data generation algorithm is given. The experiments show that the proposed methods compare favorably to the state-of-the-art by improving clustering accuracy and the speed of convergence. We also observe that the currently most popular K-means++ initialization behaves like the random one in the very high-dimensional cases.

show abstract

Section: Methodsmentioning

confidence: 99%

Improving Scalable K-Means++

2020

Self Cite

View full text Add to dashboard Cite

show abstract

“…These benchmark sets are synthetic datasets, suggested for use when testing any algorithm dealing with clustering spherical or Gaussian data. Here, we restrict ourselves to the benchmarks with at most 20 clusters, because the interpretation and knowledge discovery from the clustering results with a large number of prototypes might become tedious [6,49]. Therefore, the number of clusters was also tested with K = 2 − 25.…”

Section: Methodsmentioning

confidence: 99%

“…Hierarchical clustering constructs a tree structure from data to present layers of clustering results, but because of the pairwise distance matrix requirement, the basic form of the method is not scalable to a large volume of data [5]. Moreover, many clustering algorithms, including hierarchical clustering, can produce clusters of arbitrary shapes in the data space, which might be difficult to interpret for knowledge discovery [6].…”

Section: Introductionmentioning

confidence: 99%

Comparison of Internal Clustering Validation Indices for Prototype-Based Clustering

2017

Self Cite

View full text Add to dashboard Cite

Abstract:Clustering is an unsupervised machine learning and pattern recognition method. In general, in addition to revealing hidden groups of similar observations and clusters, their number needs to be determined. Internal clustering validation indices estimate this number without any external information. The purpose of this article is to evaluate, empirically, characteristics of a representative set of internal clustering validation indices with many datasets. The prototype-based clustering framework includes multiple, classical and robust, statistical estimates of cluster location so that the overall setting of the paper is novel. General observations on the quality of validation indices and on the behavior of different variants of clustering algorithms will be given.

show abstract

“…The SSE values were computed with formula (1) for the whole data. Finally, statistical comparison between the methods was performed with the nonparametric Kruskal-Wallis test [37], [38], since in most of the cases the clustering errors were not normally distributed. The significance level was set to 0.05.…”

Section: Methodsmentioning

confidence: 99%

Scalable Initialization Methods for Large-Scale Clustering

Hämäläinen¹,

Kärkkäinen²,

Rossi³

2020

Preprint

Self Cite

View full text Add to dashboard Cite

In this work, two new initialization methods for K-means clustering are proposed. Both proposals are based on applying a divide-and-conquer approach for the K-means type of an initialization strategy. The second proposal also utilizes multiple lower-dimensional subspaces produced by the random projection method for the initialization. The proposed methods are scalable and can be run in parallel, which make them suitable for initializing large-scale problems. In the experiments, comparison of the proposed methods to the K-means++ and K-means methods is conducted using an extensive set of reference and synthetic large-scale datasets. Concerning the latter, a novel high-dimensional clustering data generation algorithm is given. The experiments show that the proposed methods compare favorably to the state-of-the-art. We also observe that the currently most popular K-means++ initialization behaves like the random one in the very high-dimensional cases.

show abstract

Feature Ranking of Large, Robust, and Weighted Clustering Result

Cited by 9 publications

References 17 publications

Improving Scalable K-Means++

Improving Scalable K-Means++

Comparison of Internal Clustering Validation Indices for Prototype-Based Clustering

Scalable Initialization Methods for Large-Scale Clustering

Contact Info

Product

Resources

About