Bias-Corrected Bayesian Classification with Selected Features

• Activity Number: 604
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: International Society for Bayesian Analysis (ISBA)
• Author(s): Longhai Li*+
• Keywords: optimistic bias ; high-dimensional classification ; feature selection ; PAM ; DLDA

Abstract

Class prediction based on high-dimensional features has received a great deal of attention in many areas. For example, biologists are interested in using microarray gene expression profiles for diagnosis or prognosis of a certain disease (eg, cancer). For computational and other reasons, it is necessary to select a subset of features before fitting a statistical model, by looking at how strongly the features are related to the response. However, such feature selection procedure will result in overconfident predictive probabilities for future cases. We develop a hierarchical Bayesian classification method that can correct for this feature selection bias. Our method (called BCBCSF) uses the partial information from the feature selection procedure, in addition to the retained features, to form a bias-corrected posterior distribution of certain hyperparameters that control the signal-noise ratio of the data set. We take MCMC approach to infer the model parameters. Our simulation studies and tests with two real microarray data sets related to complex human diseases show that BCBCSF predicts better than two widely used high-dimensional classification methods --- PAM and DLDA.