MDL-based attribute models in naive Bayes classification
Petri Myllymäki
In: Second Workshop on Information Theoretic Methods in Science and Engineering, 17-19 August, 2009, Tampere, Finland.

Abstract

When classifying objects with Naive Bayes classifiers, we are faced with the problem of how to handle continuous attributes. Common solutions to this problem are discretizing, or assuming the data to be normally distributed. In this paper we take a different approach and instead model the class-specific attribute distributions of Na¨ıve Bayes classifiers with MDL-optimal histogram density functions. We present experimental results, comparing MDL-optimal histograms to Gaussian distributions and histograms learned with other methods.

PDF - Requires Adobe Acrobat Reader or other PDF viewer.

[Edit]