Integrating Case-Based Reasoning in Multi-Criteria Decision Support Systems

An important focus in current research on decision support systems (DSS) is the design of flexible environments to facilitate and support learning about the problem domain by the user. This research uses case-based reasoning to present a symbiotic DSS in which both the user and the DSS learn from each other. The user learns from the DSS (from stored prior problem solutions) and the system learns from the user (by observing current problem solving behaviors). The specific context of our research is the class of DSS used for supporting multi-criteria decision making (MCDM). Our ideas are being implemented in a prototype extension of the Triple C [2] MCDSS. 1 Multi-Criteria Decision Support Systems Multi-Criteria Decision Support Systems (MCDSS) are interactive, computer-based systems helping decision-makers (individuals and/or groups) to solve various semistructured and unstructured problems involving multiple attributes, objectives and goals. A recent survey of MCDSS [9] describes the wide applicability of these systems and indicates their importance within the DSS research field (see also [12], [5] and [23]). Typical problems addressed by MCDSS can be characterized as being composed of four main components: a user (an individual decision-maker or a group), a (not necessarily explicit) preference structure, a dynamic, finite set of alternatives (potential actions), and a set of criteria to be considered in the evaluation of alternatives (see Figure 1). In order to provide support during the different phases of a decision-making process, the majority of current MCDSS rely on one (or more) of the methodologies developed within the MCDM (Multi Criteria Decision Making) research field (see e.g. [13, 28]). Such methodologies address multi-criteria decision making as a problem which can be mapped onto a (generally static) formal model, and then solved using suitable mathematical techniques (see for instance the Expert Choice system based on Saaty's Analytical Hierarchy Process [22], the family of methods Electre developed by Roy [21], as well as a wide range of systems based on Multi Attribute Utility Theory [25]). In addition to mathematical models, MCDSS typically include heuristics or rule-based reasoning [15,14], and databases, as illustrated in Figure 1. Mathematical techniques developed in the MCDM field have had a major impact on the design of the early MCDSS. As a result, the majority of current MCDSS are merely computerized versions of MCDM techniques (as described above).