A Bayesian Approach to Concept Drift

Stephen H. Bach and Marcus A. Maloof

To cope with concept drift, we placed a probability distribution over the location of the most-recent drift point. We used Bayesian model comparison to update this distribution from the predictions of models trained on blocks of consecutive observations and pruned potential drift points with low probability. We compare our approach to a non-probabilistic method for drift and a probabilistic method for change-point detection. In our experiments, our approach generally yielded improved accuracy and/or speed over these other methods.

Paper available in PDF.

Bach, S. H., & Maloof, M.A. (2010). A Bayesian approach to concept drift. Advances in Neural Information Processing Systems 23, 123–135. Red Hook, NY: Curran Associates.

@incollection{bach.nips.10,
  author = "Bach, S. H. and Maloof, M. A.",
  title = "A {B}ayesian approach to concept drift",
  booktitle = "{Advances in Neural Information Processing Systems 23}", 
  year = 2010,
  pages = "127--135",
  url = "http://books.nips.cc/nips23.html",
  publisher = "Curran Associates",
  address = "Red Hook, NY",
  note = "Proceedings of the 24th Annual Conference on Neural Information
     Processing Systems", 
  annote = {
  }}