Analysing Sensitivity Data from Probabilistic Networks

With the advance of efficient algorithms for sen­ sitivity analysis of probabilistic networks, study­ ing the sensitivities revealed by real-life net­ works is becoming feasible. As the amount of data yielded by an analysis of even a moderately­ sized network is already overwhelming, effective methods for extracting relevant information from these data are called for. One such method is to study the derivatives of the sensitivity func­ tions yielded, to identify the parameters that upon variation are expected to have a large effect on a probability of interest. We further propose to build upon the concept of admissible deviation, which captures the extent to which a parameter can be varied without inducing a change in the most likely outcome. We illustrate these con­ cepts by means of a sensitivity analysis of a real­ life probabilistic network in the field of oncology.