Bayesian Network Classifiers

  • Authors:
  • Nir Friedman;Dan Geiger;Moises Goldszmidt

  • Affiliations:
  • Computer Science Division, 387 Soda Hall, University of California, Berkeley, CA 94720. E-mail: nir@cs.berkeley.edu;Computer Science Department, Technion, Haifa, Israel, 32000. E-mail: dang@cs.technion.ac.il;SRI International, 333 Ravenswood Ave., Menlo Park, CA 94025. E-mail: moises@erg.sri.com

  • Venue:
  • Machine Learning - Special issue on learning with probabilistic representations
  • Year:
  • 1997

Quantified Score

Hi-index 0.06

Visualization

Abstract

Recent work in supervised learning has shown that a surprisinglysimple Bayesian classifier with strong assumptions of independence amongfeatures, called naive Bayes, is competitive withstate-of-the-art classifiers such as C4.5. This fact raises the question ofwhether a classifier with less restrictive assumptions can perform evenbetter. In this paper we evaluate approaches for inducing classifiers fromdata, based on the theory of learning Bayesian networks. These networks are factored representations ofprobability distributions that generalize the naive Bayesian classifier andexplicitly represent statements about independence. Among these approacheswe single out a method we call Tree Augmented Naive Bayes (TAN), which outperforms naive Bayes, yet at the same timemaintains the computational simplicity (no search involved) and robustnessthat characterize naive Bayes. We experimentally tested these approaches,using problems from the University of California at Irvine repository, and compared them to C4.5, naive Bayes, and wrapper methods for featureselection.