A variable-selection heuristic for K-means clustering

Abstract

cluster analysis, K-means partitioning, variable selection, heuristics,

Variable Selection and Outlier Detection for Automated K-means Clustering

“…The Automated K-means clustering procedure consists of three processes: (i) automatically calculating the cluster number and initial cluster center whenever a new variable is added, (ii) identifying outliers for each cluster depending on used variables, (iii) selecting variables defining cluster structure in a forward manner. To select variables, we applied VS-KM (variable-selection heuristic for K-means clustering) procedure (Brusco and Cradit, 2001). To identify outliers, we used a hybrid approach combining a clustering based approach and distance based approach.…”

“…In clustering analysis, it has been frequently observed that only a limited subset of variables is valuable to defined the cluster structure (Brusco and Cradit, 2001). Furthermore, the incorporation of masking variables which do not define cluster structure may complicate or obscure the recovery of cluster structure during hierarchical or nonhierarchical cluster analysis (Milligan, 1980(Milligan, , 1989Fowlkes and Mallows, 1983;Brusco and Cradit, 2001).…”

“…Furthermore, the incorporation of masking variables which do not define cluster structure may complicate or obscure the recovery of cluster structure during hierarchical or nonhierarchical cluster analysis (Milligan, 1980(Milligan, , 1989Fowlkes and Mallows, 1983;Brusco and Cradit, 2001). For the general approaches to identify masking variables in cluster analysis, see Gnanadesikan et al (1995) and Brusco and Cradit (2001).…”

“…Furthermore, the incorporation of masking variables which do not define cluster structure may complicate or obscure the recovery of cluster structure during hierarchical or nonhierarchical cluster analysis (Milligan, 1980(Milligan, , 1989Fowlkes and Mallows, 1983;Brusco and Cradit, 2001). For the general approaches to identify masking variables in cluster analysis, see Gnanadesikan et al (1995) and Brusco and Cradit (2001). For the variable selection in K-means clustering, Carmone et al (1999) proposed a graphical variableselection procedure, named HINoV (heuristic identification of noisy variables) based on the adjusted Rand (1971) index of Hubert and Arabie (1985).…”

James–Stein shrinkage to improve k-means cluster analysis

“…This trimming method was further developed by Cuesta-Albertos et al (2008) and García-Escudero et al (2009). In addition, many authors have proposed implementations of K-means incorporating variable selection (e.g., Brusco and Cradit, 2001) or variable weighting (e.g., DeSarbo et al, 1984;Makarenkov and Legendre, 2001). Steinley (2006) provides an excellent comprehensive review of the k-means method, including the properties and a multitude of variations thereof.…”