Information Technology Reference
In-Depth Information
Evaluating Case Selection Algorithms for Analogical
Reasoning Systems
Eduardo Lupiani, Jose M. Juarez, Fernando Jimenez, and Jose Palma
Computer Science Faculty - Universidad de Murcia - Spain
{
elupiani,jmjuarez,fernan,jtpalma
}
@um.es
Abstract.
An essential issue for developing analogical reasoning systems (such
as Case-Based Reasoning systems) is to build the case memory by selecting reg-
isters from an external database. This issue is called case selection and the liter-
ature provides a wealth of algorithms to deal with it. For any particular domain,
to choose the case selection algorithm is a critical decision on the system design.
Despite some algorithms obtain good results, a specific algorithms evaluation is
needed. Most of the efforts done in this line focus on the number of registers
selected and providing a simple evaluation of the system obtained. In some do-
mains, however, the system must fulfil certain constraints related to accuracy or
efficiency. For instance, in the medical field, specificity and sensitivity are crit-
ical values for some tests. In order to partially solve this problem, we propose
an evaluation methodology to obtain the best case selection method for a given
memory case. In order to demonstrate the usefulness of this methodology, we
present new case selection algorithms based on evolutionary multi-objective op-
timization. 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.
1
Introduction
The cognitive process of reasoning by analogy comprehends a sort of different tasks
such as obtaining abstract knowledge from instances or to solve new problems based
on previous experiences [32]. One the one hand, the abstraction issue is studied, from
a computational point of view, in the instance-based learning algorithms [3] or in more
specific problems, such as temporal abstraction. On the other hand, analogical reason-
ing systems (ARSs) focus on solving a problem by comparing prior problem-solving
episodes [8,12], stored in a case memory.
Over the years, ARS have become a well known field and have demonstrated to be
essential in domains such as industry [20] or medicine [18]. Some of the most extended
approaches are the K-NN family of algorithms and the Case-based Reasoning (CBR),
a methodology to develop ARSs by providing a 4-step guided cycle [28], called CBR
cycle [1].
A solid building process and a good evaluation of the knowledge base are essential in
the practical application of ARS. Building processes mainly focus on defining (1) how
This study was partially financed by the Spanish MEC through projects TIN2009-14372-C03-
01 and PET2007-0033 and the regional project SENECA 15277/PI/10.
Search WWH ::
Custom Search