Decision-Theoritic Modelling: Decision support systems
From DSL
The principles of decision-analytic decision support, implemented in GeNIe and SMILE can be applied in practical decision support systems (DSSs). In fact, Decision Systems Laboratory has developed and is currently developing several such probabilistic DSSs in which GeNIe plays the role of a developer's environment and SMILE plays the role of the reasoning engine. A decision support system based on SMILE can be equipped with a dedicated user interface.
Probabilistic DSSs, applicable to problems involving classification, prediction, and diagnosis, are a new generation of systems that are capable of modeling any real-world decision problem using theoretically sound and practically invaluable methods of probability theory and decision theory. Based on graphical representation of the problem structure, these systems allow for combining expert opinions with frequency data, gather, manage, and process information to arrive at intelligent solutions.
Probabilistic DSSs are based on a philosophically different principle than rule-based expert systems. While the latter attempt to model the reasoning of a human expert, the former use an axiomatic theory to perform computation. The soundness of probability theory provides a clear advantage over rule-based systems that usually represent uncertainty in an ad-hoc manner, such as using certainty factors, leading to under-responsiveness or over-responsiveness to evidence and possibly incorrect conclusions.
Probabilistic DSSs are applicable in many domains, among others in medicine (e.g., diagnosis, therapy planning), banking (e.g., credit authorization, fraud detection), insurance (e.g., risk analysis, fraud detection), military (e.g., target detection and prioritization, battle damage assessment, campaign planning), engineering (e.g., process control, machine and process diagnosis), and business (e.g., strategic planning, risk analysis, market analysis).
An example DSS developed using GeNIe and SMILE is the medical diagnostic system Hepar II (Onisko et al. 1999, 2000). The system aids physicians in diagnosis of liver disorders. The structure of the model, currently consisting of almost 100 variables, has been elicited from physician experts, while its numerical parameters have been learned from a database of patient cases.
The Hepar II system is equipped with a simple dedicated user interface that allows for entering various observations such as symptoms and results of medical tests and displays the probability distribution over various possible disorders in the order of most to least likely.
The system is currently used both as a diagnostic aid and a training tool.
