Obtaining well calibrated probabilities using Bayesian binning

Naeini MP, Cooper GF, Hauskrecht M. Obtaining well calibrated probabilities using Bayesian binning. In: Proceedings of the Conference of the Association for the Advancement of Artificial Intelligence (2015) Jan;2015:2901-2907. PMID: 25927013 PMC4410090.

Learning probabilistic predictive models that are well calibrated is critical for many prediction and decision-making tasksinartificialintelligence.Inthispaperwepresentanew non-parametric calibration method called Bayesian Binning intoQuantiles(BBQ)whichaddresseskeylimitationsofexisting calibration methods. The method post processes the output of a binary classification algorithm; thus, it can be readily combined with many existing classification algorithms. The method is computationally tractable, and empirically accurate, as evidenced by the set of experiments reported here on both real and simulated datasets.

Publication Year: 
2015
Publication Credits: 
Naeini MP, Cooper GF, Hauskrecht M.
AttachmentSize
PDF icon Naeini.pdf355.22 KB
^