Abstract Details
Activity Number:
|
644
|
Type:
|
Contributed
|
Date/Time:
|
Thursday, August 8, 2013 : 8:30 AM to 10:20 AM
|
Sponsor:
|
Section on Nonparametric Statistics
|
Abstract - #308918 |
Title:
|
Correcting the Bias of Variable Selections in Cost-Sensitive Learning
|
Author(s):
|
Hongjuan Liu*+ and Bei Zhou
|
Companies:
|
and Johnson & Johnson Pharmaceutical R&D
|
Keywords:
|
Classification Tree ;
Variable Selection ;
Imbalanced Data ;
Generalized Gini Criteria ;
Permutation ;
Koziol's Combinational Approach
|
Abstract:
|
A selection bias towards covariates with many possible splits or missing values is a fundamental problem in the recursive binary partitioning tree. The variable selection bias seriously affects the interpretability of tree-structured regression models. Unbiased procedures have been suggested for some cases. However, most the algorithms assume or expect balanced class distributions. In the imbalanced data, the asymmetric sample sizes in the two classes render the approaches either rather burdensome or inapplicable to use. To correct the bias of variable selection, we investigated the algorithms for the evaluation of distribution of the maximally selected generalized Gini criteria that enables the fast computation for the imbalanced data. Moreover, we demonstrate that the cost function plays an important role in the variable selection; in particular, achieve a higher probability of the informative variable detection when the cost function of incorrectly classifying the minority class is increased if the minority class shows smaller variation than the majority class.
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2013 program
|
2013 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.
The views expressed here are those of the individual authors and not necessarily those of the JSM sponsors, their officers, or their staff.
Copyright © American Statistical Association.