Adaptive Learning Algorithm for Bayesian Networks Based on Kernel Mixtures Distributions

Abstract

Bayesian networks (BN) are a powerful tool for modelling multivariate random variables. However, when applied, for example, for industrial projects, problems arise because it does not adapt the existing learning and inference algorithms to real data, since real data are usually
represented as multivariate random variables with non-Gaussian distribution. This article discusses structural and parametric learning problems in Bayesian networks from data which has non-Gaussian distribution and non-linear relations. We propose an algorithm based on the use of mixtures of Gaussian distributions to solve a problem when the joint normality assumption is not confirmed. A feature of the algorithm is that the mixture distributions are adopted for the data right at the stage of structural learning and are used further in the process of parameters learning. Experiments have been run on both synthetic datasets and real-world data and have shown gains in modelling quality.