Case-Based Reasoning in the Health Sciences: Foundations and Research Directions - Computational Intelligence in Healthcare: Advanced Methodologies - page 135

Table 2. Sample memory and case management classifications

Code

Memory Organization Case Re presentation Flag

10 Flat T Time Series

20 Hierarchical A Text

20.1 Decision Tree M Microarray

20.2 Concept Lattice V Attribute/Values

20.3 Conceptual Clustering Tree N Plans

30 Network Memory Structures Flag

40 Inverted Index G Ground Cases

Case Re presentation Flag P Prototypes

I Images L Clusters

S Signals O Concepts

how they are organized in memory for retrieval purposes and more (see Table
2). As a result, it is made up of more than one code. The first part of the code
represents the format of the cases. The primary types are images, signals, mass
spectrometry, microarray, time series data and regular attribute/value pairs,
which are used by the majority of the systems. There are also flags that represent
what kinds of memory structures the CBR system uses, such as ground cases (G),
prototypical cases (P), clusters (L), or concepts (O). Lastly, when it comes to
memory management, there are potentially an infinite number of possibilities,
some of which may never have been used before. The main types, however,
represent how the memory is organized, whether it is flat or hierarchical, what
kind of hierarchical structure, such as decision tree, concept lattice, conceptual
clustering tree, or others.

Reasoning. This category regroups the inferential aspects of the CBR. Clas-
sically, retrieve, reuse, revise, and retain have been described. Nevertheless, re-
searchers have often added many more aspects to the inferences, such that it is
best to keep this category open to important variations (see Table 3). Each of
these parts of the reasoning cycle can be hierarchically refined so that a tree is
formed here also.

System Design. The construction of the CBR system specifies what technolo-
gies it uses. This area of classification may not seem intuitive at first, but upon
the examination of CBR systems it can be seen that many use a combination of
technologies, not just case-based reasoning. The most common technology used
in conjunction with CBR is rule-based reasoning; however some systems com-
bine CBR with information retrieval, data mining, or other artificial intelligence
methods. See Table 4 for an example of different possible construction classifica-
tions. If the construction of the system does use additional technologies, a flag
should be appended to the end of the code to denote whether the case-based
reasoning is executed separately. Also, an additional flag is used to designate
CBR's role in the system, whether primary, secondary, or equivalent.