A Study of Different Families of Fusion Functions for Combining Classifiers in the One-vs-One Strategy
In this work we study the usage of different families of fusion functions for combining classifiers in a multiple classifier system of One-vs-One (OVO) classifiers. OVO is a decomposition strategy used to deal with multi-class classification problems, where the original multi-class problem is divided into as many problems as pair of classes. In a multiple classifier system, classifiers coming from different paradigms such as support vector machines, rule induction algorithms or decision trees are combined. In the literature, several works have addressed the usage of classifier selection methods for these kinds of systems, where the best classifier for each pair of classes is selected. In this work, we look at the problem from a different perspective aiming at analyzing the behavior of different families of fusion functions to combine the classifiers. In fact, a multiple classifier system of OVO classifiers can be seen as a multi-expert decision making problem. In this context, for the fusion functions depending on weights or fuzzy measures, we propose to obtain these parameters from data. Backed-up by a thorough experimental analysis we show that the fusion function to be considered is a key factor in the system. Moreover, those based on weights or fuzzy measures can allow one to better model the aggregation problem.
KeywordsAggregations Fusion functions Classification One-vs-One Multiple classifier system
This work was supported in part by the Spanish Ministry of Science and Technology under Project TIN2016-77356-P (AEI/FEDER, UE).
- 1.Aha, D.W., Kibler, D., Albert, M.K.: Instance-based learning algorithms. Mach. Learn. 6, 37–66 (1991)Google Scholar
- 2.Alcalá-Fdez, J., Fernandez, A., Luengo, J., Derrac, J., García, S., Sánchez, L., Herrera, F.: KEEL data-mining software tool: data set repository, integration of algorithms and experimental analysis framework. J. Multiple-Valued Logic Soft Comput. 17(2–3), 255–287 (2011)Google Scholar
- 10.Elkano, M., Galar, M., Sanz, J., Fernandez, A., Barrenechea, E., Herrera, F., Bustince, H.: Enhancing multi-class classification in farc-hd fuzzy classifier: on the synergy between n-dimensional overlap functions and decomposition strategies. IEEE Trans. Fuzzy Syst. 23(5), 1562–1580 (2015)CrossRefGoogle Scholar
- 15.Grabisch, M.: A new algorithm for identifying fuzzy measures and its application to pattern recognition. In: International Joint Conference of the 4th IEEE International Conference on Fuzzy Systems and the 2nd International Fuzzy Engineering Symposium, pp. 145–150 (1995)Google Scholar
- 24.Quinlan, J.R.: C45: Programs for Machine Learning, 1st edn. Morgan Kaufmann Publishers, San Mateo (1993)Google Scholar