CLUSTERING OF COUNT DATA USING POISSON DISTRIBUTION
Advances and Applications in Statistics · 2025
DOI: 10.17654/0972361725053 ↗Kurz gefasst
Fachartikel in der Zeitschrift Advances and Applications in Statistics, 2025. Verfasst von S. Krishnamoorthy und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Cluster analysis is often used to identify homogeneous groups within complex datasets, particularly when traditional distance-based methods struggle with high-dimensional or skewed data. In this study, we propose a model-based clustering approach for count data using a finite mixture of Poisson distributions. The model accounts for overdispersion and skewness, with parameters estimated via the expectation-maximization (EM) algorithm. Information criteria such as AIC and BIC are employed for model selection. A key novelty of this work lies in applying Poisson mixture models to a large-scale health survey dataset, specifically the behavioral risk factor surveillance system (BRFSS), treating BMI as discrete count data. The proposed method is also benchmarked against lognormal mixture models, demonstrating superior performance in terms of misclassification rate and adjusted rand index (ARI). Additionally, the impact of initialization strategies on EM convergence is examined using both real and simulated datasets. Results confirm that Poisson mixture-based clustering offers a more effective and interpretable solution for count data than traditional approaches.
Zitieren
APA
Krishnamoorthy, S., & Jaganathan, B. (2025). CLUSTERING OF COUNT DATA USING POISSON DISTRIBUTION. Advances and Applications in Statistics. https://doi.org/10.17654/0972361725053
Vancouver
Krishnamoorthy S, Jaganathan B. CLUSTERING OF COUNT DATA USING POISSON DISTRIBUTION. Advances and Applications in Statistics. 2025. doi:10.17654/0972361725053
BibTeX
@article{Krishnamoorthy2025,
author = {Krishnamoorthy, S. and Jaganathan, B.},
title = {CLUSTERING OF COUNT DATA USING POISSON DISTRIBUTION},
journal = {Advances and Applications in Statistics},
year = {2025},
doi = {10.17654/0972361725053}
}
EUVIMED