Clustering is one of the core problems in machine learning. Many clustering algorithms aim to partition data along a single dimension. This approach may become inappropriate when data has higher dimension and is multifaceted. This paper introduces a class of mixture models with multiple dimensions called pouch latent tree models. We use them to perform cluster analysis on a data set consisting of 75 development indicators for 133 countries. We further propose a method that guides the selection of clustering variables due to the existence of multiple latent variables. The analysis results demonstrate that some interesting clusterings of countries can be obtained from mixture models with multiple dimensions but not those with single dimensions. Copyright © 2017 Springer International Publishing AG.
|Title of host publication||Advances in neural networks - ISNN 2017: 14th International Symposium, ISNN 2017, Sapporo, Hakodate, and Muroran, Hokkaido, Japan, June 21-26, 2017, Proceedings, Part I|
|Editors||Fengyu CONG , Andrew LEUNG, Qinglai WEI|
|Place of Publication||Cham|
|Publication status||Published - 2017|
CitationPoon, L. K. M. (2017). Clustering with multidimensional mixture models: Analysis on world development indicators. In F. Cong, A. Leung, & Q. Wei (Eds), Advances in neural networks - ISNN 2017: 14th International Symposium, ISNN 2017, Sapporo, Hakodate, and Muroran, Hokkaido, Japan, June 21-26, 2017, Proceedings, Part I (pp. 153-160). Cham: Springer.
- Multidimensional clustering
- Pouch latent tree models
- Mixture models
- World development indicators
- Clustering variables selection