Lecture by Prof. Sanjib Basu in Faculty Hall @ 4pm

Location: faculty Hall

TITLE: Feature Selection: Statistical Methods with Application to Biomarker Selection

SPEAKER: Professor Sanjib Basu
Paul Levy and Virginia F. Tomasek
Professor of Biostatistics, and
the Director of the Center for Biostatistical Development at University of Illinois,
Chicago (UIC).


Feature Selection, also known as variable selection, is the process of selecting a subset of relevant features/variables/predictors. Methods for feature selection is an ever-expanding area of research. Filter type feature selection methods typically process features one at a time whereas wrapper methods consider subsets of features, and evaluate model selection criteria for each subset. Embedded methods inherently perform variable selection as part of the learning process and include various regularization methods (such as the Lasso). We propose a Bayesian method for variable selection. While this is a general statistical approach, the motivation comes from the problem of selecting relevant biomarkers associated with (a) lymph node metastasis and (b) improved survival from Non-Small Cell Lung Cancer. Both of these are non-linear models which adds substantial complexity to the feature selection problem.

About the Speaker:

Sanjib Basu is the Paul Levy and Virginia F. Tomasek Professor of Biostatistics, and the Director of the Center for Biostatistical Development at University of Illinois at Chicago (UIC). Before joining UIC, he was a Presidential Research Professor and the Director of the Division of Statistics at Northern Illinois University. Dr. Basu’s scholarship has been nationally and internationally recognized by Fellowship and elected membership from the *American Statistical Association* and the *International Statistical Institute* and by principal investigator grant awards from US National Institute of Health, National Science Foundation, and Pharmaceutical Industry. He is currently serving on the Editorial board of two prestigious Statistics journals and served on multiple Editorial boards in the past. His scholarly research in statistical methodology include Bayesian modeling, inference, feature selection with applications in biomedicine, cancer survival and epidemiology and reliability. Dr. Basu also has a strong and sustained record of collaborative research with biomedical researchers, medical oncologists, thoracic surgeons and others on clinical, genetic, proteomic, epigenetic features and their impact on outcomes.

Director will preside