|
Activity Number:
|
382
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #305760 |
|
Title:
|
Classification by Ensembles from Random Partitions of High-Dimensional Genomic Data
|
|
Author(s):
|
Hojin Moon*+ and Hongshik Ahn and James J. Chen and Ralph L. Kodell
|
|
Companies:
|
U.S. Food and Drug Administration and Stony Brook University and U.S. Food and Drug Administration and U.S. Food and Drug Administration
|
|
Address:
|
3900 NCTR Road, Jefferson, AR, 72079,
|
|
Keywords:
|
class prediction ; classification tree ; cross validation ; majority voting ; risk profiling
|
|
Abstract:
|
A robust classification algorithm is developed based on ensembles of classifiers, with each classifier constructed from a different set of predictors determined by a random partition of the entire set of predictors. The proposed method combines the results of multiple classifiers to achieve a substantially improved prediction compared to the optimal single classifier. By combining classifiers built from each subspace of the predictors, our algorithm achieves a huge computational advantage in tackling the growing problem of dimensionality. We investigate the performance of our method compared to widely used classification methods using real data. Our classification algorithm has many applications, including the early detection of disease in apparently disease-free individuals based on multiple biomarkers and the assignment of patients to drug therapies based on genomic and other profiles.
|
