Databases Reference
In-Depth Information
(First Order Inductive Learner) is an inductive learning algorithm for
generating CARs developed by Quinlan and Cameron-Jones in 1993. This
algorithm was later developed by Yin and Han to produce the PRM (Pre-
dictive Rule Mining) CAR generation algorithm. PRM was then further
developed, by Yin and Han in 2003 to produce CPAR (Classification based
on Predictive Association Rules).
Case Satisfaction Approaches
Regardless of which particular methodology is used to build it, a classifier is
usually presented as an ordered CAR list
R
. In [15] Coenen and Leng sum-
marised three case satisfaction approaches that have been employed in differ-
ent CARM algorithms for utilising the resulting classifier to classify “unseen”
data. These three case satisfaction approaches are itemised as follows (given
a particular case):
•
Best First Rule:
Select the first “best” rule that satisfies the given case
according to some ordering imposed on
. The ordering can be de-
fined according to many different ordering schemes, including: (1) CSA
(Confidence-Support-size of Antecedent) - combinations of confidence,
support and size of antecedent, with confidence being the most significant
factor (used in CBA, TFPC and the early stages of processing of CMAR);
(2) WRA (Weighted Relative Accuracy) - which reflects a number of rule
“interestingness” measures as proposed in [35]; (3) Laplace Accuracy -
as used in PRM and CPAR; (4)
χ
2
Testing -
χ
2
values as used, in part,
in CMAR; (5) ACS (size of Antecedent-Confidence-Support) - an alter-
native to CSA that considers the size of the rule antecedent as the most
significant factor; etc.
R
•
Best
K
Rules:
Select the first “best
K
” rules (in this chapter we denote
by
k
as mentioned above) that satisfy the given case and then select a
rule according to some averaging process as used for example, in CPAR.
The term “best” in this case is defined according to an imposed ordering
of the form described in
Best First Rule
.
K
•
All Rules:
Collect all rules in the classifier that satisfy the given case and
then evaluate this collection to identify a class. One well-known evaluation
method in this category is WCS (Weighted
χ
2
) testing as used in CMAR.
Rule Ordering Approaches
As noted in the previous section five existing rule ordering mechanisms are
identified to support the “best first rule” case satisfaction strategy. Each can
be further separated into two stages: (1) a rule weighting stage where each
R
j
∈R
is labeled with a weighting score that represents the significance of
R
j
indicates a single class
c
i
; and (2) a rule re-ordering stage, which sorts the
original
R
in a descending manner, based on the score assigned in stage (1), of
