Project Risk Management by a Probabilistic Expert System

Abstract

Efficient applications of expert systems to project risk management problems are seldom, if not unusual. In this paper we overcome this lack by using the probabilistic expert system shell SPIRIT. The rule-based shell ’s power in conditioning, inference and reasoning under incomplete information will work well on risk estimation and classification. A key characteristic of SPIRIT is the possibility to integrate project objectives into the risk management model. So known dependencies between risk variables can be modelled by the user if known beforehand, whereas hidden dependencies might be detected by the proper system. Because of the novelty of projects they suffer from incomplete information and it is this incompleteness which SPIRIT handles at high information fidelity. Furthermore undirected inference is possible, due to the undirected graphical structure in which knowledge is acquired and processed. So, in an early-state risk management situation - where the final model in terms of certain variables and/or their respective dependencies is not yet available - preliminary risk analyses and even recommendations for adequate risk treatment measures are possible, too. A middle size product developement example, including 12 binary variables and 34 rules, shows the inferential power of SPIRIT.