Clustering in the Presence of Scatter

Abstract

A new methodology is proposed for clustering datasets in the presence of scattered observations. Scattered observations are defined as unlike any other, so traditional approaches that force them into groups can lead to erroneous conclusions. Our suggested approach is a scheme which, under assumption of homogeneous spherical clusters, iteratively builds cores around their centers and groups points within each core while identifying points outside as scatter. In the absence of scatter, the algorithm reduces to k-means. We also provide methodology to initialize the algorithm and to estimate the number of clusters in the dataset. Results in experimental situations show excellent performance, especially when clusters are elliptically symmetric. The methodology is applied to the analysis of the United States Environmental Protection Agency's Toxic Release Inventory reports on industrial releases of mercury for the year 2000. KeywordsBayes Information Criterion, biweight estimator, exact-c-separation, k-clips, MCLUST, methylmercury, tight clustering Disciplines Statistics and Probability CommentsThis is the peer reviewed version of the following article: Maitra,R., and Ramler, I. Clustering in the presence of scatter. A new methodology is proposed for clustering datasets in the presence of scattered observations. Scattered observations are defined as unlike any other, so traditional approaches that force them into groups can lead to erroneous conclusions. Our suggested approach is a scheme which, under assumption of homogeneous spherical clusters, iteratively builds cores around their centers and groups points within each core while identifying points outside as scatter. In the absence of scatter, the algorithm reduces to k-means. We also provide methodology to initialize the algorithm and to estimate the number of clusters in the dataset. Results in experimental situations show excellent performance, especially when clusters are elliptically symmetric. The methodology is applied to the analysis of the United States Environmental Protection Agency (EPA)'s Toxic Release Inventory (TRI) reports on industrial releases of mercury for the year 2000.

An efficientk‐means‐type algorithm for clustering datasets with incomplete records

“…The k-means algorithm does not make distributional assumptions but may be cast in a semi-parametric framework [12], [41]. We now show that the following holds even without the Gaussian distributional assumptions that underlie Result 1:…”

A Bayesian mixture model to quantify parameters of spatial clustering

“…These drawbacks have been tackled by more recent methods. Maitra and Ramler (2009), e.g., proposed a generalization of the k-means algorithm that explicitly considers scattered points. Some sophisticated grouping algorithms were proposed that only require specifying, e.g., a maximal cluster size (Scharl and Leisch, 2006), a minimal cluster size (Manley M. Schäfer et al / Computational Statistics and Data Analysis ( ) -et al, 2008, relying on point trajectories over time) or both a minimal cluster size and an effective maximal cluster radius (Ester et al, 1996;Ankerst et al, 1999).…”

Integrative sparse $K$-means with overlapping group lasso in genomic applications for disease subtype discovery

“…Popular methods include hierarchical clustering [Eisen et al (1998)], K -means [Dudoit and Fridlyand (2002)], mixture model-based approaches [Xie, Pan and Shen (2008); McLachlan, Bean and Peel (2002)] and nonparametric approaches [Qin (2006)], for analysis of single transcriptomic study. Resampling and ensemble methods have been used to improve stability of the clustering analysis [Kim et al (2009); Swift et al (2004)] or to pursue tight clusters by leaving scattered samples that are different from major clusters [Tseng (2007); Tseng and Wong (2005); Maitra and Ramler (2009)]. Witten and Tibshirani (2010) proposed a sparse K -means algorithm that can effectively select gene features and perform sample clustering simultaneously.…”