Using CBR as Design Methodology for Developing Adaptable Decision Support Systems - Efficient Decision Support Systems: Practice and Challenges from Current to Future - page 125

Fig. 2. Methods and techniques commonly used for implementing each stage of a typical
CBR system
As the core of CBR systems is its memory, cases should therefore accurately represent
both problems and their solutions. In this context, cases may be deleted if they are found
to produce inaccurate solutions, they may be merged together to create more generalised
solutions, and they may be modified, over time, through the experience gained in
producing improved solutions. If an attempt to solve a problem fails and it is possible to
identify the reason for the failure, then this information should also be stored in order to
avoid the same mistake in the future. This corresponds to a common learning strategy
employed in human problem-solving. Rather than creating general relationships between
problem descriptors and conclusions, as is the case with rule-based reasoning, or relying
on general knowledge of the problem domain, CBR systems are able to utilise the specific
knowledge of previously experienced, concrete problem situations. A CBR system
provides an incremental learning process because each time a problem is solved, a new
experience is retained, thus making it available for future reuse.
As design methodology adequate for constructing DSS, case-based reasoning can be used by
itself or as part of another intelligent or conventional computing system. From the last years,
there has been an increasing interest in the possibility of integrating different AI techniques
with the goal of constructing more powerful and accurate hybrid systems. In this context,
the work of Soucek (1991) established the IRIS ( Integration of Reasoning, Informing and
Serving ) classification with the goal of facilitating the efficient design of intelligent systems.
In the same line, the work of Medsker & Bailey (1992) proposed five integration models
based on symbolic and connectionistic systems. Finally, the work of Bezdek (1994)
suggested the CIC ( Computational Intelligence Classification ) schema, an interesting
classification guidelines for cataloguing hybrid AI systems.
Therefore, given their hybrid nature, CBR systems can be easily combined with other
alternatives in order to construct robust decision support systems. These approaches include
their successful hybridization with expert systems (Vo & Macchion, 1993; Rissland et al.
1993; Medsker, 1995), fuzzy logic (Xu, 1994; Gui, 1993; Dubois et al. 1997), genetic algorithms
(Louis et al. 1993; Oppacher & Deugo, 1991), qualitative reasoning (Navinchandra et al.
1991), constraint satisfaction systems (Maher & Zhang, 1993; Hinrichs, 1992), artificial neural
networks (Thrift, 1989; Lim et al. 1991; Liu & Yan, 1997; Corchado et al. 2001) and bayesian