A Cost-Sensitive Paradigm for Multiclass to Binary Decomposition Schemes
An established technique to face a multiclass categorization problem is to reduce it into a set of two-class problems. To this aim, the main decomposition schemes employed are one vs. one, one vs. all and Error Correcting Output Coding. A point not yet considered in the research is how to apply these methods to a cost-sensitive classification that represents a significant aspect in many real problems. In this paper we propose a novel method which, starting from the cost matrix for the multi-class problem and from the code matrix employed, extracts a cost matrix for each of the binary subproblems induced by the coding matrix. In this way, it is possible to tune the single two-class classifier according to the cost matrix obtained and achieve an output from all the dichotomizers which takes into account the requirements of the original multi-class cost matrix. To evaluate the effectiveness of the method, a large number of tests has been performed on real data sets. The experiments results have shown a significant improvement in terms of classification cost, specially when using the ECOC scheme.
KeywordsFalse Negative Rate Decomposition Scheme Cost Matrix True Negative Rate Code Matrix
- 5.Tortorella, F.: An Empirical Comparison of In-Learning and Post-Learning Optimization Schemes for Tuning the Support Vector Machines in Cost-Sensitive Applications. In: Proc. 12th Int. Conf. on Image Anal. and Proc., pp. 560–565. IEEE Computer Society Press, Los Alamitos (2003)CrossRefGoogle Scholar
- 6.Blake, C., Keogh, E., Merz, C.J.: UCI Repository of Machine Learning Databases (1998), www.ics.uci.edu/~mlearn/MLRepository.html
- 7.Joachims, T.: Making Large-Scale SVM Learning Practical. In: Schölkopf, B., Burges, C.J.C., Smola, A.J. (eds.) Advances in Kernel Methods, pp. 169–184. MIT Press, Cambridge (1999)Google Scholar
- 8.NIST/SEMATECH e-Handbook of Statistical Methods (2003), http://www.itl.nist.gov/div898/handbook/
- 9.Margineantu, D.D., Dietterich, T.G.: Bootstrap Methods for the Cost-Sensitive Evaluation of Classifiers. In: Proc. Int. Conf. Machine Learning ICML 2000, pp. 582–590 (2000)Google Scholar