RUcore Resource Object
RUcore Resource Object
PDF-1(1.09 MB)
TitleStatistical modeling and inference for multiple temporal or spatial cluster detection
NameSun, Qiankun (author), Xie, Minge (chair), Naus, Joseph (co-chair), Kolassa, John (internal member), Zhao, Jun (outside member), Rutgers University, Graduate School - New Brunswick,
Degree Date2008-10
Date Created2008
SubjectStatistics and Biostatistics,
Cluster analysis
DescriptionThis thesis develops a latent modeling framework and likelihood based inference tool to detect multiple temporal or spatial clusters.
Cluster detection is important to researchers from various fields. Practical applications include: biological studies of DNA sequencing, environmental researches, epidemiological studies and surveillance for biological terrorism. The traditional scan statistics procedures have technical difficulties to detect multiple clusters of varying sizes. Some Bayesian approaches have to limit
the potential clusters in cell divisions. A recently proposed stepwise regression method tends to be inefficient in some cases. We utilize some probability distributions to model the latent clusters and mimic the sample data generation process. With model selection techniques, we can obtain an optimal number of total potential clusters. Based on a Monte-Carlo EM algorithm and likelihood based inference, we are able to estimate the associated model parameters, detect significant clusters and identify their locations and sizes. Compared with other procedures, this new approach is intuitive and simple. It is also more efficient and flexible for further extensions.
Cluster detection is important to researchers from various fields. Practical applications include: biological studies of DNA sequencing, environmental researches, epidemiological studies and surveillance for biological terrorism. The traditional scan statistics procedures have technical difficulties to detect multiple clusters of varying sizes. Some Bayesian approaches have to limit
the potential clusters in cell divisions. A recently proposed stepwise regression method tends to be inefficient in some cases. We utilize some probability distributions to model the latent clusters and mimic the sample data generation process. With model selection techniques, we can obtain an optimal number of total potential clusters. Based on a Monte-Carlo EM algorithm and likelihood based inference, we are able to estimate the associated model parameters, detect significant clusters and identify their locations and sizes. Compared with other procedures, this new approach is intuitive and simple. It is also more efficient and flexible for further extensions.
NotePh.D.
NoteIncludes bibliographical references (p. 75-78)
Noteby Qiankun Sun
Genretheses
Persistent URLhttp://hdl.rutgers.edu/1782.2/rucore10001600001.ETD.000050446
Languageeng
CollectionGraduate School - New Brunswick Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.