COMPSTAT 2016: Start Registration
View Submission - COMPSTAT
Title: A model based clustering for ordered categorical data Authors:  Tadashi Imaizumi - Tama University (Japan) [presenting]
Abstract: The volumes of data matrix have been larger and larger in modern data analysis. And the clustering methods will be useful to find several homogenous group from data. However, as we also treat many categorical data such text data in a newspaper, data collected by many sensors, etc., we need to think about how to model for analyzing these categorical data. One approach will be to adopt the model-based clustering approach. The latent class models and methods will be one alternative as model based approach. The categorical variables will be represented a binary vector corresponding to a categorical value of a categorical variables. So, the data matrix with $p$ categorical variables will be a high dimensional and sparse data matrix. Though these latent class methods will be useful, but, they do not fit for data with larger sample cases or many categorical variables. And we need to develop a new model and methods of the latent class analysis or the model-based clustering for these type of data matrix. A model and method will be proposed for uncovered the classes of $n$ samples and classes of $p$ categorical variables in frame of two-way clustering.