On inclusion-driven learning of bayesian networks
The Journal of Machine Learning Research
Discriminative versus generative parameter and structure learning of Bayesian network classifiers
ICML '05 Proceedings of the 22nd international conference on Machine learning
Learning Bayesian Networks: A MAP Criterion for Joint Selection of Model Structure and Parameter
ICDM '08 Proceedings of the 2008 Eighth IEEE International Conference on Data Mining
Building classifiers using Bayesian networks
AAAI'96 Proceedings of the thirteenth national conference on Artificial intelligence - Volume 2
A multivariate discretization method for learning Bayesian networks from mixed data
UAI'98 Proceedings of the Fourteenth conference on Uncertainty in artificial intelligence
| Hi-index | 0.00 |
In Probabilistic Seismic Hazard Analysis, which has become the basis of decision making on the design of high risk facilities, one estimates the probability that ground motion caused by earthquakes exceeds a certain level at a certain site within a certain time interval. One of the most critical aspects in this context is the model for the conditional probability of ground motion given earthquake magnitude, source-site-distance and potentially additional parameters. These models are usually regression functions, including terms modelling interaction effects derived from expert knowledge. We show that the framework of Directed Graphical Models is an attractive alternative to the standard regression approach. We investigate Bayesian Networks, modelling the problem in a true multivariate way, and we look into Naive Bayes and Tree-Augmented Naive Bayes, where the target node coincides with the dependent variable in standard ground motion regression. Our approach gives rise to distribution-free learning when necessary, and we experiment with and introduce different discretization schemes to apply standard learning and inference algorithms to our problem at hand.