A knowledge-based decision support system for managerial problem recognition and diagnosis
Ata Mohamed, Nassar H
MetadataShow full item record
Current decision support systems lack in providing support for the early phases of decision making (i.e., problem finding and problem diagnosis). This research develops a conceptual model for managerial problem recognition and diagnosis, develops a system architecture, implements a prototype system based on a computer-based simulated management game, and tests the prototype with a set of problems. The conceptual model is based on Pounds' model of problem finding and Sage's model of cognitive processing. The conceptual model consists of a problem finding component (Monitor) that detects deviations in specified variables, and a problem processor. The problem processor supports routine diagnoses (that have occurred in the past) through the experiential knowledge base. New problem diagnosis is supported through several processes such as construction of causation trees, hypotheses generation, hypotheses evaluation, verification of diagnosis, and statistical verification of user's models. The elements of the full system architecture consist of a User Interface, Performance Monitor, Security System, Monitor for Problem Detection, Problem Processor, Structural Model Knowledge Base, System Dictionary, and a Process Controller. In the prototype implementation, the components Performance Monitor and Security System were not included; also the problem processor implementation did not include the component for statistical verification of the user's models. Structural modeling was used for knowledge representation since it allows representation of causal relations among variables, which is important in business problem domains. Results show the technical feasibility of the approach and confirm the view of some researchers that perhaps the domain of business problems requires more than one method of knowledge representation. Results also indicate that the quality of diagnosis depends on the quality of the model from which diagnosis is inferred. These results were discussed in relation to the type of models for which the approach is appropriate and new questions for future research were proposed.