Sensitivity Analysis in Bayesian networks with Mixtures of Truncated Base Functions

Abstract

Sensitivity analysis is a technique used to determine the robustness of the output of a mathematical model to inaccuracies in the assessments of its parameters. An existing method of sensitivity analysis for discrete Bayesian networks, where the effect of varying quantitative parameters on the output is analysed, is generalised towards a type of hybrid Bayesian network, namely the Bayesian network with Mixtures of Truncated Base Functions. The generalisation offers multiple ways of varying the parameter functions, such as by shifting and stretching, and gives multiple ways of co-varying the other parameters, where proportional co-variation is deemed best.