A Mixture-of-Experts Framework for Learning from Imbalanced Data Sets

Abstract

Re-Sampling methods are some of the different types of approaches proposed to deal with the class-imbalance problem. Although such approaches are very simple, tuning them most effectively is not an easy task. In particular, it is unclear whether oversampling is more effective than undersampling and which oversampling or undersampling rate should be used. This paper presents an experimental study of these questions and concludes that combining different expressions of the resampling approach in a mixture of experts framework is an effective solution to the tuning problem. The proposed combination scheme is evaluated on a subset of the REUTERS-21578 text collection (the 10 top categories) and is shown to be very effective when the data is drastically imbalanced.

We would like to thank Rob Holte and Chris Drummond for their useful comments. This research was funded, in part, by an NSERC grant. The work conducted in this paper was conducted at Dalhousie University.