Improving the Performance of FARC-HD in Multi-class Classification Problems Using the One-Versus-One Strategy and an Adaptation of the Inference System
In this work we study the behavior of the FARC-HD method when addressing multi-class classification problems using the One-vs-One (OVO) decomposition strategy. We will show that the confidences provided by FARC-HD (due to the use of the product in the inference process) are not suitable for this strategy. This problem implies that robust strategies like the weighted vote obtain poor results. For this reason, we propose two improvements: 1) the replacement of the product by greater aggregations whose output is independent of the number of elements to be aggregated and 2) the definition of a new aggregation strategy for the OVO methodology, which is based on the weighted vote, in which we only take into account the confidence of the predicted class in each base classifier. The experimental results show that the two proposed modifications have a positive impact on the performance of the classifier.
KeywordsClassification One-vs-One Fuzzy Rule-Based Classification Systems Aggregations
Unable to display preview. Download preview PDF.
- 1.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. Journal of Multiple-Valued Logic and Soft Computing 17(2-3), 255–287 (2011)Google Scholar
- 4.Friedman, J.H.: Another approach to polychotomous classification. Tech. rep., Department of Statistics, Stanford University (1996), http://www-stat.stanford.edu/~jhf/ftp/poly.ps.Z
- 8.Ishibuchi, H., Nakashima, T., Nii, M.: Classification and modeling with linguistic information granules: Advanced approaches to linguistic Data Mining. Springer, Berlin (2004)Google Scholar
- 11.Sheskin, D.: Handbook of parametric and nonparametric statistical procedures. Chapman & Hall/CRC (2006)Google Scholar