Intelligent model management in an actuarial consulting system
Taracad R. Sivasankaran · 1985
When multiple decision models are to be considered simultaneously, expertise is required for selecting, adapting and coordinating them. This research investigates intelligent model management for decision support systems using the example of actuarial problem solving. The contributions are threefold. First, the proposed model management component handles a very large number of models. Second, the system does not just provide a model manipulation language but includes intelligent solution planning. Finally, it offers the capability to detect and acquire missing data. A prototype Actuarial Consulting System (ACS)--currently with about 200 models--has been implemented to demonstrate the above concepts. Actuaries evaluate the risks and compute premiums for providing insurance for life contingencies such as death, disability or retirement. The ACS represents static actuarial knowledge using a graph formalism called Formula Derivation Network (FDN). Each FDN is implemented at a surface level and at an execution level, thus separating planning from computations. Dynamic knowledge is structured in terms of a Process Model that captures the general theory of problem solving in the actuarial domain. The Problem Analyzer converts a problem stated by the user into a search problem on the stored collection of FDNs. The Solution Planner plans the search process to solve this problem and the Plan Executor evaluates the sequence of models and formulas selected by the Solution Planner. Model management is implemented through the Solution Planner which employs several types of problem-solving strategies including human expert rules. A Fact Acquisition Model detects situations with insufficient information, determines the additional relevant information required for solving the problem, and acquires such information from a database or from the user.