Information Technology Reference
In-Depth Information
the cases (instances) are stored in the case memory, (2) a retrieval strategy to obtain the
most similar cases from the case memory and (3) how the solution is built from these
knowledge [13]. The evaluation of knowledge-based systems, including ARSs, depends
on their knowledge base. Unlike other approaches, such as rule-based or model-based
systems [19], in an ARS each piece of the knowledge base (cases of the case memory) is
knowledge-complete and independent from the rest. Therefore, the main issue to build
a case memory is to select which cases must be included or removed.
Two main approaches can be considered. Firstly, if the system is already established,
the case memory must be maintained . This CBR task deals with the decision of re-
moving, adding or modifying elements of the case memory [15]. Secondly, during the
building process, the case memory can be obtained by selecting registers from an ex-
ternal data base. The automatic process to carry out this selection is known as instance
selection, case selection or case mining [17,21].
The literature provides a wealth of case selection algorithms [9,7,2,22,27,29,31],
however during the ARS building process, the selection of the most suitable algorithm
is critical. Despite the impact of this process in the ARS evaluation, a few efforts have
been done in this direction.
This work focuses on the case selection task and its evaluation. To this end, we
introduce an evaluation method for case selection algorithms. In order to demonstrate
the usefulness of this methodology, we present new case selection algorithms based on
multi-objective evolutionary optimization. We compare the classical algorithms and the
multi-objective approach in order to select the most suitable case selection algorithm
according to different standard problems.
The structure of this work is as follows. Section 2 provides a review of the related
works on case selection methods and evaluation. In Section 3 we propose a evalua-
tion methodology for case selection methods. Section 4 describes the experiments and
results. Finally, conclusions and future works are described in Section 5.
2
Related Works
The present work is aimed to evaluate case selection methods. We briefly review what
kind of case selection families and which evaluation techniques have been studied in
literature.
Case selection methods can be classified either by the case selection method-
ology [30,31] or by case memory construction technique [21]. Among the case
selection methods there are four outstanding families: (i) nearest neighbour edit-
ing rules [9,7,29,22,27], (ii) instance-based [3,2,31], (iii) prototype-based [5] and
(iv) competence-based [24,26,17]. Both (i) and (ii) select a case as candidate to final
case memory in basis to K-NN classification and (iii) could additionally adapt cases or
modify them. Specially interesting methods are those based on competence techniques
(iv) since they introduce new concepts such as coverage and reachability; however
they need good understanding of domain problem. Case selection and feature selection
approaches has been very popular research issue to reduce noise in case memory and
enhance system response times respectively, however almost all studies face them sep-
arately. Multi-objetive evolutionary algorithms (MOEA) makes possible to cope both
Search WWH ::




Custom Search