Article ID: | iaor200937802 |
Country: | Germany |
Volume: | 166 |
Issue: | 1 |
Start Page Number: | 23 |
End Page Number: | 38 |
Publication Date: | Feb 2009 |
Journal: | Annals of Operations Research |
Authors: | Terlaky Tams, Peng Jiming, Jiao Tianshi |
Keywords: | datamining |
In this paper, we deal with ranking problems arising from various data mining applications where the major task is to train a rank-prediction model to assign every instance a rank. We first discuss the merits and potential disadvantages of two existing popular approaches for ranking problems: the ‘Max-Wins’ voting process based on multi-class support vector machines (SVMs) and the model based on multi-criteria decision making. We then propose a confidence voting process for ranking problems based on SVMs, which can be viewed as a combination of the SVM approach and the multi-criteria decision making model. Promising numerical experiments based on the new model are reported.