Athens University Economics and Business (AUEB)
Model-based clustering for count and mixed mode data
Abstract
dc:descriptionModel based clustering (MBC) is a common approach for modelling data with the use of finite mixtures of parametric distributions. For count data, the choice of high dimensional multivariate Poisson distribution can lead to increased computational effort. Composite likelihoods concept with the use of bi-variate marginals, can offer flexibility in estimations. In order to further reduce the time of estimation of the composite likelihood method associated parameters, in this thesis we introduce the sampling methods which can offer adequate results, especially for large data samples. When it comes to mixed data sets, the joint probability is not always easy to be found. Copulas are well known as flexible models which allow creating multivariate distributions with given marginals. Hence, they can create a wealth of multivariate models including models with different marginal distributions. The purpose of the present thesis is mainly to expand the derived so far results of using copula-based models for MBC applications. Gaussian Copula offers flexibility for description of the dependencies between different types of variables. Our aim is to further reduce computational effort arisen from the use of Gaussian copula, and the fully parametrized model we assessed, since this approach causes effort from adding different correlation matrices for every component that need to be estimated. So, the main target is to achieve parsimony in estimation with the use of appropriate techniques. Parsimonious alternatives have been proposed based on dimension reduction approaches like factor analysis or clever representations of the correlation matrices named structured correlation matrices.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Panagou, Foteini
- Πανάγου, Φωτεινή
Subjects
dc:subject × 14Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/53259