Discriminative learning of Bayesian networks via factorized conditional log-likelihood

Alexandra M. Carvalho, Teemu Teppo Roos, Arlindo L. Oliveira, Petri Myllymäki

Research output: Contribution to journalArticleScientificpeer-review

Abstract

We propose an efficient and parameter-free scoring criterion, the factorized conditional log-likelihood (ˆfCLL), for learning Bayesian network classifiers. The proposed score is an approximation of the conditional log-likelihood criterion. The approximation is devised in order to guarantee decomposability over the network structure, as well as efficient estimation of the optimal parameters, achieving the same time and space complexity as the traditional log-likelihood scoring criterion. The resulting criterion has an information-theoretic interpretation based on interaction information, which exhibits its discriminative nature. To evaluate the performance of the proposed criterion, we present an empirical comparison with state-of-the-art classifiers. Results on a large suite of benchmark data sets from the UCI repository show that ˆfCLL-trained classifiers achieve at least as good accuracy as the best compared classifiers, using significantly less computational resources.
Original languageEnglish
JournalJournal of Machine Learning Research
Volume12
Pages (from-to)2181-2210
Number of pages30
ISSN1532-4435
Publication statusPublished - 2011
MoE publication typeA1 Journal article-refereed

Fields of Science

  • 113 Computer and information sciences
  • 112 Statistics and probability

Cite this

@article{48fd9811f50d47908d36b6d5c532c581,
title = "Discriminative learning of Bayesian networks via factorized conditional log-likelihood",
abstract = "We propose an efficient and parameter-free scoring criterion, the factorized conditional log-likelihood (ˆfCLL), for learning Bayesian network classifiers. The proposed score is an approximation of the conditional log-likelihood criterion. The approximation is devised in order to guarantee decomposability over the network structure, as well as efficient estimation of the optimal parameters, achieving the same time and space complexity as the traditional log-likelihood scoring criterion. The resulting criterion has an information-theoretic interpretation based on interaction information, which exhibits its discriminative nature. To evaluate the performance of the proposed criterion, we present an empirical comparison with state-of-the-art classifiers. Results on a large suite of benchmark data sets from the UCI repository show that ˆfCLL-trained classifiers achieve at least as good accuracy as the best compared classifiers, using significantly less computational resources.",
keywords = "113 Computer and information sciences, 112 Statistics and probability",
author = "Carvalho, {Alexandra M.} and Roos, {Teemu Teppo} and Oliveira, {Arlindo L.} and Petri Myllym{\"a}ki",
year = "2011",
language = "English",
volume = "12",
pages = "2181--2210",
journal = "Journal of Machine Learning Research",
issn = "1532-4435",
publisher = "Microtome Publishing",

}

Discriminative learning of Bayesian networks via factorized conditional log-likelihood. / Carvalho, Alexandra M.; Roos, Teemu Teppo; Oliveira, Arlindo L.; Myllymäki, Petri.

In: Journal of Machine Learning Research, Vol. 12, 2011, p. 2181-2210.

Research output: Contribution to journalArticleScientificpeer-review

TY - JOUR

T1 - Discriminative learning of Bayesian networks via factorized conditional log-likelihood

AU - Carvalho, Alexandra M.

AU - Roos, Teemu Teppo

AU - Oliveira, Arlindo L.

AU - Myllymäki, Petri

PY - 2011

Y1 - 2011

N2 - We propose an efficient and parameter-free scoring criterion, the factorized conditional log-likelihood (ˆfCLL), for learning Bayesian network classifiers. The proposed score is an approximation of the conditional log-likelihood criterion. The approximation is devised in order to guarantee decomposability over the network structure, as well as efficient estimation of the optimal parameters, achieving the same time and space complexity as the traditional log-likelihood scoring criterion. The resulting criterion has an information-theoretic interpretation based on interaction information, which exhibits its discriminative nature. To evaluate the performance of the proposed criterion, we present an empirical comparison with state-of-the-art classifiers. Results on a large suite of benchmark data sets from the UCI repository show that ˆfCLL-trained classifiers achieve at least as good accuracy as the best compared classifiers, using significantly less computational resources.

AB - We propose an efficient and parameter-free scoring criterion, the factorized conditional log-likelihood (ˆfCLL), for learning Bayesian network classifiers. The proposed score is an approximation of the conditional log-likelihood criterion. The approximation is devised in order to guarantee decomposability over the network structure, as well as efficient estimation of the optimal parameters, achieving the same time and space complexity as the traditional log-likelihood scoring criterion. The resulting criterion has an information-theoretic interpretation based on interaction information, which exhibits its discriminative nature. To evaluate the performance of the proposed criterion, we present an empirical comparison with state-of-the-art classifiers. Results on a large suite of benchmark data sets from the UCI repository show that ˆfCLL-trained classifiers achieve at least as good accuracy as the best compared classifiers, using significantly less computational resources.

KW - 113 Computer and information sciences

KW - 112 Statistics and probability

M3 - Article

VL - 12

SP - 2181

EP - 2210

JO - Journal of Machine Learning Research

JF - Journal of Machine Learning Research

SN - 1532-4435

ER -