A New Instance-Based Label Ranking Approach Using the Mallows Model

In this paper, we introduce a new instance-based approach to the label ranking problem. This approach is based on a probability model on rankings which is known as the Mallows model in statistics. Probabilistic modeling provides the basis for a theoretically sound prediction procedure in the form of maximum likelihood estimation. Moreover, it allows for complementing predictions by diverse types of statistical information, for example regarding the reliability of an estimation. Empirical experiments show that our approach is competitive to start-of-the-art methods for label ranking and performs quite well even in the case of incomplete ranking information. Key words: Instance-based learning, Label ranking, Classication, Maximum likelihood estimation

• Type: Conference
• Year: 2009
• Where: ISNN
• Authors: Weiwei Cheng, Eyke Hüllermeier
• Keywords: Artificial Intelligence, ISNN 2009, Label Ranking, Label Ranking Problem, Maximum Likelihood Estimation