碩士 / 逢甲大學 / 統計學系統計與精算碩士班 / 106 / The issues of model-based clustering and classification of longitudinal data have received increased attention in recent years. The finite mixtures of t linear mixed-effects model (FM-tLMM) has become one of the commonly used analytical tools for performing this work when data contain outliers. In this thesis, we propose an extended finite mixtures of t linear mixed-effects model (Extended FM-tLMM), where the categorical latent variables (component labels) are assumed to be influenced by the fixed observed covariates. As compared with the traditional FM-tLMM which assumes the mixing proportions to be fixed but unknown, our proposed Extended FM-tLMM utilizes a logistic regression model to link the relationship between prior classification probabilities and the covariates of interest. Therefore, the Extended FM-tLMM can offer more accurate estimates of model parameters and better classification performance. In order to obtain maximum likelihood estimates of model parameters, we reformulate the model into hierarchical structures and develop the alternating expectation conditional maximization (AECM) algorithm. Meanwhile, we calculate the standard errors of the parameter estimators by an information-based method. The issue of clustering individuals under the fitted model is also investigated. We utilize a real data example concerning the AIDS clinical trials and simulation studies to investigate the performance of the proposed model and compare it with the existing models in terms of model selection, parameter estimation and classification.
Identifer | oai:union.ndltd.org:TW/106FCU00336016 |
Date | January 2018 |
Creators | YANG, YU-CHEN, 楊郁成 |
Contributors | Wang, Wan-Lun, 王婉倫 |
Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
Language | zh-TW |
Detected Language | English |
Type | 學位論文 ; thesis |
Format | 76 |
Page generated in 0.0101 seconds