TY - JOUR
T1 - An asymptotic analysis of some expert fusion methods
AU - Chen, Dechang
AU - Cheng, Xiuzhen
PY - 2001/6
Y1 - 2001/6
N2 - We study the asymptotic behavior of three classifier combination methods for two-class classification: Average, median, and majority vote. Assuming that the estimates of the posterior probability given by individual classifiers constitute a sample from a distribution, we show that as the number of individual classifiers becomes large, median and majority will produce the same result but average may yield a completely different decision if the distribution is not symmetric.
AB - We study the asymptotic behavior of three classifier combination methods for two-class classification: Average, median, and majority vote. Assuming that the estimates of the posterior probability given by individual classifiers constitute a sample from a distribution, we show that as the number of individual classifiers becomes large, median and majority will produce the same result but average may yield a completely different decision if the distribution is not symmetric.
KW - Classifier combination
KW - Error rate
KW - Majority vote
KW - Order statistics
UR - http://www.scopus.com/inward/record.url?scp=0035368916&partnerID=8YFLogxK
U2 - 10.1016/S0167-8655(01)00031-9
DO - 10.1016/S0167-8655(01)00031-9
M3 - Article
AN - SCOPUS:0035368916
SN - 0167-8655
VL - 22
SP - 901
EP - 904
JO - Pattern Recognition Letters
JF - Pattern Recognition Letters
IS - 8
ER -