Weighted Voting as Approximate MAP Prediction in Pairwise Classification

Weighted voting is the commonly used strategy for combining predictions in pairwise classification. Even though it shows excellent performance in practice, it is often criticized for lacking a sound theoretical justification. In this paper, we study the problem of combining predictions within a formal framework of label ranking. In this framework, we derive a generalized voting strategy in which predictions are properly adapted according to the strength of the corresponding base classifiers, and which is optimal in the sense of yielding a MAP prediction. Then, we show that weighted voting yields a good approximation of this MAP prediction. This theoretical argument in favor of weighted voting as a quasi-optimal aggregation strategy is further corroborated by empirical evidence from experiments with real and synthetic data sets.

• Type: Conference
• Year: 2008
• Where: LWA
• Authors: Eyke Hüllermeier, Stijn Vanderlooy
• Keywords: Generalized Voting Strategy, LWA 2008, MAP Prediction, Software Engineering, Weighted Voting