An Ensemble Approach for High-Dimensional Data Clustering Using Random-Sample Adaptive Subspaces and Reliability-Weighted GMM Method
IOSR Journal of Computer Engineering · 2026-05 · 28 (5) · 01-06
DOI: 10.9790/0661-2805010106 ↗Kurz gefasst
Fachartikel in der Zeitschrift IOSR Journal of Computer Engineering, 2026-05. Verfasst von P.M.Chandni P.M.Chandni und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Background
Cluster analysis of high dimensional data poses various challenges including problems associated with the curse of dimensionality, presence of irrelevant attributes, and high computational costs. Conventional GMM based techniques do not perform well with data having thousands of attributes and small number of observations, e.g., Coil20, GLIOMA, and Lung datasets. Materials and Methods: The proposed algorithm includes three steps as follows: 1) RS-ASG: Random-Sample Adaptive Subspace Generation - the normalized data set is randomly divided into partitions and adaptive subspaces are generated through informative feature ranking for dimensionality reduction and diversity. 2) RASGMM: Reliability Aware Subspace GMM - GMM technique is used on each partition-subspace combination and the optimum number of components is selected with BIC criterion. Each GMM model is assigned with reliability weight depending on its clustering performance. 3) CPPC: Cross-Partition Probabilistic Consensus - clustering results from various partitions are matched and combined together with reliability aware probabilistic consensus method. The clustering performance is measured with ACC and NMI.
Results
Results on Coil20, GLIOMA, and Lung datasets indicate that RS-AW-SGMM-CE has a larger NMI and ACC than GMMs, GMM after subspace learning and SubGMM-CE.
Conclusion
The proposed RS-AW-SGMM-CE framework presented in this work solves such issues related to high-dimensional clustering tasks with the integration of adaptive subspace generation, reliability-based weighting and consensus fusion. The experimental results verify that RS-AW-SGMM-CE is a scalable and effective method for large-scale high-dimensional datasets with significantly increased accuracy, stability and reliability in clustering
Zitieren
APA
P.M.Chandni, P., & Hridya, G. H. G. (2026). An Ensemble Approach for High-Dimensional Data Clustering Using Random-Sample Adaptive Subspaces and Reliability-Weighted GMM Method. IOSR Journal of Computer Engineering, 28(5), 01-06. https://doi.org/10.9790/0661-2805010106
Vancouver
P.M.Chandni P, Hridya GHG. An Ensemble Approach for High-Dimensional Data Clustering Using Random-Sample Adaptive Subspaces and Reliability-Weighted GMM Method. IOSR Journal of Computer Engineering. 2026;28(5):01-06. doi:10.9790/0661-2805010106
BibTeX
@article{PMChandni2026,
author = {P.M.Chandni, P.M.Chandni and Hridya, G. Hridya G.},
title = {An Ensemble Approach for High-Dimensional Data Clustering Using Random-Sample Adaptive Subspaces and Reliability-Weighted GMM Method},
journal = {IOSR Journal of Computer Engineering},
year = {2026},
volume = {28},
number = {5},
pages = {01--06},
doi = {10.9790/0661-2805010106}
}
EUVIMED