Benchmarking BNT Inference Engines using an Early Warning System

Abstract

Bayesian Networks (BN) are systematic probabilistic reasoning models for visualization and inspection of interrelated data due to their ability to capture uncertain knowledge in an intuitive and efficient manner. They can be applied for various tasks such as decision making, reliability analysis, reasoning, etc. There are numerous toolboxes that support and simplify the usage of Bayesian Networks such as Bayes Net Toolbox (BNT) which is one of the most academically used and comprehensive BN toolbox. Inference algorithms supported by BNT differ in terms of performance. Therefore it is crucial to establish a criterion for selecting appropriate inference algorithms. The goal of this paper is to establish a benchmark of inference algorithms supported by BNT on a 49 node sized Early Warning System model.