Decision templates for multiple classi"er fusion: an experimental comparison
Ludmila Ilieva Kuncheva, James C. Bezdek, Robert P. W. Duin · 2001
Multiple classier fusion may generate more accurate classication than each of the constituent classiers. Fusion is often based on xed combination rules like the product and average. Only under strict probabilistic conditions can these rules be justied. We present here a simple rule for adapting the class combiner to the application. c decision templates (one per class) are estimated with the same training set that is used for the set of classiers. These templates are then matched to the decision prole of new incoming objects by some similarity measure. We compare 11 versions of our model with 14 other techniques for classier fusion on the Satimage and Phoneme datasets from the database ELENA. Our results show that decision templates based on integral type measures of similarity are superior to the other schemes on both data sets. ( 2000 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.