This is the program for the 2010 Joint Statistical Meetings in Vancouver, British Columbia.
Abstract Details
Activity Number:
|
589
|
Type:
|
Contributed
|
Date/Time:
|
Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Section on Statistical Learning and Data Mining
|
Abstract - #308358 |
Title:
|
Correlated Component Regression: A Prediction/Classification Methodology for Possibly Many Features
|
Author(s):
|
Jay Magidson*+
|
Companies:
|
Statistical Innovations Inc.
|
Address:
|
7 Stevens Terrace, Arlington, MA, 02478, United States
|
Keywords:
|
high dimensional data ;
feature selection ;
log-linear models ;
K-component model ;
sequential independence ;
event history
|
Abstract:
|
A new ensemble regression technique, called Correlated Component Regression (CCR), is proposed that involves sequential application of the Naïve Bayes rule. The general approach yields K correlated components, weights associated with a first component providing direct effects for the features, and each additional component providing improved prediction. When at least one suppressor variable is available, good prediction is generally attainable with K<5, even with the number of predictors P relatively small. An optional step-down variable selection procedure provides a sparse solution, reducing the number of features to P* < P and improving predictive performance outside the sample.
Simulation results suggest that when predictors include one or more suppressor variables, CCR models predict and select better than popular sparse penalized regression and sparse PLS regression approaches.
|
The address information is for the authors that have a + after their name.
Authors who are presenting talks have a * after their name.
Back to the full JSM 2010 program
|
2010 JSM Online Program Home
For information, contact jsm@amstat.org or phone (888) 231-3473.
If you have questions about the Continuing Education program, please contact the Education Department.