Abstract
We consider an open issue in the design of pattern classifiers, i.e., choosing between the best classifier among a given ensemble, and combining all the available ones using a trainable fusion rule. While the latter choice can in principle outperform the former, their actual effectiveness is affected by small sample size problems. This raises the need of investigating under which conditions one choice is better than the other one. We provide a first contribution, by deriving an analytical expressions of the expected error probability of best classifier selection, and by comparing it with the one of a well known linear fusion rule, implemented with the Fisher linear discriminant.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
Similar content being viewed by others
References
Duin, R.P.W.: The Combining Classifier: To Train Or Not To Train. In: Proc. 16th Int. Con. Pattern Recognition, vol. II, pp. 765–770 (2002)
Fumera, G., Roli, F.: A Theoretical and Experimental Analysis of Linear Combiners for Multiple Classifier Systems. IEEE Trans. Pattern Analysis and Machine Intelligence 27(6), 942–956 (2005)
Rao, N.S.V.: On fusers that perform better than best sensor. IEEE Trans. Pattern Analysis and Machine Intelligence 23(8), 904–909 (2001)
Raudys, S.: On the amount of a priori information in designing the classification algorithm. Engineering Cybernetics N4, 168–174 (1972) (in Russian)
Raudys, S.: Statistical and Neural Classifiers. Springer, London (2001)
Raudys, S.: Experts’ Boasting in Trainable Fusion Rules. IEEE Trans. Pattern Analysis and Machine Intelligence 25(9), 1178–1182 (2001)
Raudys, S.: Trainable fusion rules. I. Large sample size case. Neural Networks 19, 1506–1516 (2006); Trainable fusion rules. II. Small sample-size effects. Neural Networks 19, 1517–1527 (2006)
Raudys, S.: Portfolio of automated trading systems: Complexity and learning set size issues. IEEE Trans. Neural Networks Learning Systems 24(3), 448–459 (2013)
Takeshita, T., Toriwaki, J.: Experimental study of performance of pattern classifiers and the size of design samples. Patt. Rec. Lett. 16, 307–312 (1995)
Vapnik, V.N.: The Nature of Statistical Learning Theory. Springer, Berlin (1995)
Wyman, F., Young, D., Turner, D.: A comparison of asymptotic error rate expansions for the sample linear discriminant function. Patt. Rec. 23, 775–783 (1990)
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2014 Springer International Publishing Switzerland
About this paper
Cite this paper
Raudys, S., Fumera, G., Raudys, A., Pillai, I. (2014). Sample Size Issues in the Choice between the Best Classifier and Fusion by Trainable Combiners. In: Corchado, E., Lozano, J.A., Quintián, H., Yin, H. (eds) Intelligent Data Engineering and Automated Learning – IDEAL 2014. IDEAL 2014. Lecture Notes in Computer Science, vol 8669. Springer, Cham. https://doi.org/10.1007/978-3-319-10840-7_6
Download citation
DOI: https://doi.org/10.1007/978-3-319-10840-7_6
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-10839-1
Online ISBN: 978-3-319-10840-7
eBook Packages: Computer ScienceComputer Science (R0)