SDP Public Curated Content

Sensitivity Analysis for Decision Making: From Decision Trees to Artificial Intelligence

SDP Joint Webinar with DAS & SRA

Featuring: Emanuele Borgonovo (Bocconi University)
Moderated by: Robin Dillon-Merrill (DAS-INFORMS)

Abstract: 

Sensitivity Analysis is an essential part of the Decision Analysis Cycle. It plays a central
role in making decision support models more transparent. In parallel with the transformation of the models we use to support decisions, sensitivity analysis methods are rapidly evolving. Starting from the notion of information value, we will move through old and new techniques. For variable importance we will go from Tornado Diagrams to Global Methods through Shapley Values, for direction of change we will compare the familiar Spiderplots with the machine learning Partial Dependence Plots. We provide practical examples and analyze their use in applications. We will focus on how sensitivity analysis helps decision makers to interrogate models, and analysts to make clearer recommendations to decision makers.