Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

McGill University

$12,000.00

Model selection and statistical inference in mixture distributions and hidden markov (regression) models

FY 2010-11

Summary
Mixture distributions provide a flexible and useful statistical model for many problems arising in scientific research. They are typically used to analyze data that arise from a heterogeneous population or process. When the scientific background is insufficient to determine the appropriate order of a finite mixture model in an application, the number of components or order of the mixture is subject to inference. In marketing or genetics applications, it is often natural to study a general class of models: finite mixture regression (FMR) models, in which different subsets of explanatory variables are significant in different components of the mixture model. In market segmentation studies, FMR models allow researchers to investigate the phenomenon that different features of a certain product may appeal to different consumers. Identifying these features together with the number of segmentations provides information on which products are likely to be successful before being introduced to the market. Both variable selection and order selection are the key statistical issues in such applications of FMR models. In genetics, the problem becomes even more complex where the number of potential explanatory variables is in the order of hundreds or thousands. Even under a sparsity assumption on the FMR model, it is still a great statistical challenge to find tens of significant variables out of hundreds of thousands when the sample size is relativley small. In applications such as imaging, econometrics, and medical sciences, in addition to unobserved heterogeneity in data, there is also correlation in data. For example, data are spatial or made over time. Hidden Markov models (HMM) and also hidden Markov regression models (HMRM), as generalization of mixture models, are commonly used for analyzing data of this nature. Selection of the number of hidden states and/or variable selection is a task of major importance in applications of HMM and HMRM. The main theme of the proposal is to propose computationally efficient and statistically optimal solutions to the aforementioned model selection problems.
Program
Discovery Grants Program - Individual
Location
Quebec, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
386578-2010
Name nom
Khalili, Abbas
Department département
Mathematics and Statistics
Organization id
61
Institution établissement
McGill University
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2010
Competition year année de concours
2010
Award amount
12000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
Discovery grants programs
Group fr
Programmes de subventions à la découverte
Committee code
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1205
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Mathematical sciences
Area of application fr
Sciences mathématiques
Research subject code
3009
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Parametric inference
Research subject fr
Inférence paramétrique
Partie
1
Nb partie
5
Application title
Model selection and statistical inference in mixture distributions and hidden markov (regression) models
Keyword
NSERC
Application summary
Mixture distributions provide a flexible and useful statistical model for many problems arising in scientific research. They are typically used to analyze data that arise from a heterogeneous population or process. When the scientific background is insufficient to determine the appropriate order of a finite mixture model in an application, the number of components or order of the mixture is subject to inference. In marketing or genetics applications, it is often natural to study a general class of models: finite mixture regression (FMR) models, in which different subsets of explanatory variables are significant in different components of the mixture model. In market segmentation studies, FMR models allow researchers to investigate the phenomenon that different features of a certain product may appeal to different consumers. Identifying these features together with the number of segmentations provides information on which products are likely to be successful before being introduced to the market. Both variable selection and order selection are the key statistical issues in such applications of FMR models. In genetics, the problem becomes even more complex where the number of potential explanatory variables is in the order of hundreds or thousands. Even under a sparsity assumption on the FMR model, it is still a great statistical challenge to find tens of significant variables out of hundreds of thousands when the sample size is relativley small. In applications such as imaging, econometrics, and medical sciences, in addition to unobserved heterogeneity in data, there is also correlation in data. For example, data are spatial or made over time. Hidden Markov models (HMM) and also hidden Markov regression models (HMRM), as generalization of mixture models, are commonly used for analyzing data of this nature. Selection of the number of hidden states and/or variable selection is a task of major importance in applications of HMM and HMRM. The main theme of the proposal is to propose computationally efficient and statistically optimal solutions to the aforementioned model selection problems.