Summary: Cost-conscious classifier ensembles
Cigdem Demir *,1
, Ethem Alpaydin
Department of Computer Engineering, Bogazici University, Istanbul TR-34342, Turkey
Received 6 July 2004; received in revised form 16 March 2005
Available online 23 May 2005
Communicated by K. Tumer
Ensemble methods improve the classification accuracy at the expense of testing complexity, resulting in increased
computational costs in real-world applications. Developing a utility-based framework, we construct two novel cost-con-
scious ensembles; the first one determines a subset of classifiers and the second dynamically selects a single classifier.
Both ensembles successfully switch between classifiers according to the accuracy-cost trade-off of an application.
Ó 2005 Elsevier B.V. All rights reserved.
Keywords: Ensemble techniques; Utility theory; Computational cost; Voting; Selection
Different types of costs in machine learning have
been extensively investigated to date (Turney,
2000). Among them are the cost of feature extrac-
tion, the cost of misclassification errors, and the cost
of computation. The cost of feature extraction arises