Computational Agents in Complex Decision Support Systems

Abstract

The article introduces a general approach to decision making in complex systems and architecture for agent-based decision support systems (DSS). The approach contributes to decentralization and local decision making within a standard work flow. The architecture embodies the logics of the decision developing work flow and is virtually organized as a layered structure, where each level is oriented to solve one of the three following goals: data retrieval, fusion and pre-processing; data mining and evaluation; and, decision making, alerting, solutions and predictions generation. In order to test our approach, we have designed and implemented an agent-based DSS, which deals with an environmental issue. The system calculates the impacts imposed by the pollutants on the morbidity, creates models and makes forecasts by permitting to try possible ways of situation change. We discuss some used data mining techniques, namely, methods and tools for classification, function approximation, association search, difference analysis, and others. Besides, to generate sets of administrative solutions, we develop decision creation and selection work flows, which are formed and then selected in accordance with the maximum of possible positive effect and evaluated by external and internal criteria. To conclude, we show that our system provides all the necessary steps for standard decision making procedure by using computational agents. We use so much traditional data mining techniques, as well as other hybrid methods, with respect to data nature. The combination of different tools enables gaining in quality and precision of the reached models, and, hence, in the recommendations that are based on these models. The received dependencies of interconnections and associations between the factors and dependent variables help correcting recommendations and avoiding errors.