Using Association Rules for Classification from Databases Having Class Label Ambiguities: A Belief Theoretic Method - Data Mining: Foundations and Practice - page 535

Tabl e 4 . Pruned rule set

Outlook Humidity Windy Decision Confidence

sunny MEDIUM Play 1.0000

sunny FALSE Play 1.0000

LOW TRUE Play 1.0000

LOW Play 1.0000

rainy FALSE Don't play 1.0000

HIGH FALSE Don't play 1.0000

overcast HIGH Don't play 0.6667

overcast FALSE Don't play 0.6667

rainy HIGH TRUE Θ C 1.0000

sunny HIGH TRUE Θ C 1.0000

overcast HIGH TRUE Θ C 0.5000

Tabl e 5 . Final rule set generated at the conclusion of the rule refinement stage

Outlook Humidity Windy Decision Confidence

sunny MEDIUM Play 1.0000

sunny FALSE Play 1.0000

LOW TRUE Play 1.0000

LOW Play 1.0000

rainy FALSE Don't play 1.0000

HIGH FALSE Don't play 1.0000

overcast HIGH Don't play 0.6667

overcast FALSE Don't play 0.6667

rainy HIGH TRUE Θ C 1.0000

sunny HIGH TRUE Θ C 1.0000

overcast HIGH TRUE Θ C 0.5000

overcast HIGH TRUE Don't play 1.0000

These experiments use several databases from the UCI data repository [3]
and data sets collected from the airport terminal simulation platform devel-
oped at the Distributed Decision Environments (DDE) Laboratory at the
Department of Electrical and Computer Engineering, University of Miami.
The databases contain both numerical and nominal attributes. All the classi-
fication accuracies are presented with 10-fold sub-sampling where the training
and testing data sets are constructed by taking 70% and 30% of the data
instances in the database respectively. The training data set was used to gen-
erate the classification rules and the testing data set was used to evaluate its
performance.