Using CBR as Design Methodology for Developing Adaptable Decision Support Systems - Efficient Decision Support Systems: Practice and Challenges from Current to Future - page 126

networks (Shinmori, 1998; Aamodt y Langseth, 1998; Dingsøyr, 1998; Friese, 1999; Langseth
et al., 1999) among others.
Furthermore, case-based reasoning can be a particularly appropriate problem-solving
strategy when the knowledge required to formulate a rule-based model of the domain is
difficult to obtain, or when the number or complexity of rules relating to the problem
domain is too large for conventional knowledge acquisition methods. In this sense,
according to the work of Aamodt and Plaza (1994) there are five different types of CBR
systems, and although they share similar features, each of them is more appropriate for a
particular type of problem: ( i ) exemplar based reasoning (EBR), ( ii ) instance based reasoning
(IBR), ( iii ) memory-based reasoning (MBR), ( iv ) analogy-based reasoning (ABR) and ( v )
typical case-based reasoning (CBR).
EBR systems are especially suitable for classification tasks, in which the category of the most
similar past case becomes the solution to the new problem. The set of classes directly
constitutes the collection of possible solutions applied without modification. IBR systems are
a specialization of exemplar-based reasoning for solving problems in which the instances
(cases) are usually very simple (e.g. feature vectors). These systems can be completely
automated with no user intervention in their whole life cycle. MBR systems supplement
previous approaches with the capacity of parallel processing computation. ABR systems are
particularly applicable for solving new problems based on past cases from a different
domain. In order to properly work, it should be possible to transfer the solution of a source
analogue situation to the present target problem. In typical CBR systems cases are assumed
to present a certain complexity level (in terms of their internal structure and the information
contained), therefore some modification is needed in order to adapt retrieved solutions
when applied to a different problem solving context.
3. Practical applications
The decision support systems covered in this chapter come from four different research
areas: industrial planning, biomedical domain, oceanographic forecasting and anti-spam
filtering. All the implemented applications are fully designed following the CBR paradigm
in order to empower their adaptability and accuracy for solving new problems in their
respective fields.
For each domain, we first introduce the target problem to be solved together with the main
aspects surrounding each particular situation. A clear description of the representation used
for defining the case base is presented and the internal architecture governing each system is
explained in detail.
3.1 Industrial planning
Production scheduling is one of the most important functions in a production company. As
a consequence, in recent decades various methods have been proposed for the modelling
and solution of particular scheduling problems (Akyol & Bayhan, 2007). In the particular
case of cooperative poultry farms, the accurate coordination of centralized feed supply
(production and distribution) between scattered farms is of utmost importance for both the
main feed manufacturer and participating farmers.
In such a situation, some key aspects involving the main participants need to be taken into
consideration, for example ( i ) the feed production plant has a limited production and
storage capacity, ( ii ) the plant manufactures several types of feed that can create resource