Title | ||
---|---|---|
A New Approach to Learning Bayesian Network Classifiers from Data: Using Observed Statistical Frequencies |
Abstract | ||
---|---|---|
A new approach to learning Bayesian Network classifiers, called the OSF-Classification method is presented. Most approaches try and fit the network to data. The usual method is to incorporate the log-likelihood score. Analysis of this score shows that a good score could still lead to bad classifiers. The new approach, rather than trying to fit the network to data, scores the network according to its classification error. One major assumption is made, in that the parameters of the learnt network are the observed statistical frequencies of the data. This method is shown to perform well against standard Non-Bayesian learning methods. |
Year | DOI | Venue |
---|---|---|
1998 | 10.1007/BFb0057463 | LECTURE NOTES IN ARTIFICIAL INTELLIGENCE |
Keywords | Field | DocType |
bayesian learning | Brier score,Knowledge representation and reasoning,Pattern recognition,Computer science,Minimum description length,Frequency,Greedy algorithm,Bayesian network,Artificial intelligence,Probabilistic logic,Machine learning,Knowledge acquisition | Conference |
Volume | ISSN | Citations |
1480 | 0302-9743 | 0 |
PageRank | References | Authors |
0.34 | 9 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Tarkan Tahseen | 1 | 0 | 0.34 |
Duncan Fyfe Gillies | 2 | 97 | 17.86 |