Clustering noisy data in a reduced dimension space via multivariate regression trees

Smyth, Christine; Coomans, Danny; Everingham, Yvette

doi:10.1016/j.patcog.2005.09.003

Cited by 14 publications

(8 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This indicates that the actual AAMRTs and MRTPCs were similar. The similarity between the results of AAMRTs and MRTPCs has been noted elsewhere [6]. The misclassifications using these two tree types were fairly stable.…”

Section: B Golub Datasetsupporting

confidence: 79%

“…To assess the sensitivity of the results to varying values of these parameters, we ran Algorithm 1 with either (1,5,10) minimum terminal node size and either (2,4,6) terminal nodes. There were 3*3=9 choices of parameters and an ensemble of trees was grown for each choice.…”

Section: Methodsmentioning

confidence: 99%

“…Factor analysis attempts to explain the variables by assuming that they can be generated as a linear combination of q unobservable common factors (usually q << P) plus a unique factor [15]. We use either the principal component scores from the first q principal components or the factor scores from the q factors as the response variables of the tree [6]. The clustering is obtained in the reduced dimension space as q is less than P .…”

Section: Yimentioning

confidence: 99%

“…If clustering in a low dimensional setting, the explanatory variables are replicated as the response variables (autoassociative multivariate regression tree), and the clusters are found in the entire variable space. If clustering in a high dimensional setting, the dimension is first reduced using principal components analysis or factor analysis, and the resulting scores are used as the response variables [6]. The response set can be made as small as desired by taking the first q principal components or factors.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Clustering Microarrays with Predictive Weighted Ensembles

Smyth

Coomans

2007

2007 IEEE Symposium on Computational Intelligence and Bioinformatics and Computational Biology

View full text Add to dashboard Cite

Cluster ensembles seek a consensus across many individual partitions and the resulting solution is usually stable. Cluster ensembles are well suited to the analysis of DNA microarrays, where the tremendous size of the dataset can thwart the discovery of stable groups.Post processing cluster ensembles, where each individual partition is weighted according to its relative accuracy improves the performance of the ensemble whilst maintaining its stability. However, weighted cluster ensembles remain relatively unexplored, primarily because there are no common means of assessing the accuracy of individual clustering solutions. This paper describes a technique of creating weighted cluster ensembles suitable for use with microarray datasets. A regression technique is used to obtain individual cluster solutions. Each solution is then weighted according to its predictive accuracy. The consensus partition is obtained using a novel modification to the traditional k-means algorithm which further enforces the predictability of the solution. An estimate of the natural number of clusters can also be obtained using the modified k-means algorithm. Furthermore, a valuable byproduct of this weighted ensemble approach is a variable importance list. The methodology is applied on two well-known microarray datasets with promising results.

show abstract

Section: B Golub Datasetsupporting

confidence: 79%

Section: Methodsmentioning

confidence: 99%

Section: Yimentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Clustering Microarrays with Predictive Weighted Ensembles

Smyth

Coomans

2007

2007 IEEE Symposium on Computational Intelligence and Bioinformatics and Computational Biology

View full text Add to dashboard Cite

show abstract

“…Also, future work could incorporate recent advances in the AAMRT methodology. This could involve using factor scores as response variables [28]. Factor scores are considered to be more robust in the large dataset setting.…”

Section: Discussionmentioning

confidence: 99%

Predictive weighting for cluster ensembles

Smyth

Coomans

2007

Journal of Chemometrics

Self Cite

View full text Add to dashboard Cite

An ensemble of regression models predicts by taking a weighted average of the predictions made by individual models. Calculating the weights such that they reflect the accuracy of individual models (post processing the ensemble) has been shown to increase the ensemble's accuracy. However, post processing cluster ensembles has not received as much attention because of the inherent difficulty in assessing the accuracy of an individual cluster model. By enforcing the notion that clusters must be 'predictable', this paper suggests a means of implicitly post processing cluster ensembles by drawing analogies with regression post processing techniques. The product of the post processing procedure is an intelligently weighted co-occurrence matrix. A new technique, similarity-based k-means (SBK), is developed to split this matrix into clusters. The results using three real life datasets underpinned by chemical and biological phenomena show that splitting an intelligently weighted co-occurrence matrix gives accuracy that approaches supervised classification methods.

show abstract