Learning Bayesian network parameters from small data sets: application of Noisy-OR gates

摘要: Abstract Existing data sets of cases can significantly reduce the knowledge engineering effort required to parameterize Bayesian networks. Unfortunately, when a set is small, many conditioning are represented by too few or no records and they do not offer sufficient basis for learning conditional probability distributions. We propose method that uses Noisy-OR gates requirements in probabilities. test our on H epar II , model diagnosis liver disorders, whose parameters extracted from real, small patient records. Diagnostic accuracy multiple-disorder enhanced with was 6.7% better than plain 14.3% single-disorder model.