|
Activity Number:
|
260
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2008 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #301429 |
|
Title:
|
Selective Partially Augmented Naïve Bayes with Model Averaging
|
|
Author(s):
|
Yuan Yuan*+ and Jun S. Liu+
|
|
Companies:
|
Harvard University and Harvard University
|
|
Address:
|
, , MA, 02138, , Cambridge, MA, 02138,
|
|
Keywords:
|
Naive Bayes ; Bayes network ; Variable selection
|
|
Abstract:
|
Naïve Bayes is a classifier that assumes conditional independence of all features (covariates) given the class label. Although this assumption is unrealistic, the naïve Bayes classifier is competitive with some state-of-the-art classifiers. Many methods have been developed to relax this independence assumption to achieve better classification accuracy. Here we propose a Selective Partially Augmented Naïve Bayes model (SPAN) which considers the joint distribution of small groups of selected features in an augmented network structure. Feature and network structure selection is achieved by applying Markov chain Monte Carlo methods. Instead of inferring one single model, Bayesian model averaging is used to make prediction (classification). Our model is especially useful in problems with large number of covariates such as DNA marker association studies.
|
