A method for Bayesian meta-inference applying multiple regressions
Carlos Eduardo Bognar, Osamu Saotome · International Conference on Systems · 2008
This paper presents a method based on multiple regression models to select algorithms for the inference tasks in Bayesian networks. The method may be applied when exact and approximate schemes are used to perform inferences. Multiple characterizations of Bayesian networks and prediction models are considered to select the algorithm that will provide the least relative error in future inferences. Logistic regression model is applied to determine when exact algorithms may be used for specific tasks. The prediction models of approximate inference algorithms are created by multiple regression analysis, based on simulation data using Variable Elimination, Gibbs Sampling and Stratified Simulation algorithms. Experimental analyses compare some alternative approaches and show better results when multivariate analysis is applied.