Discretization - Data Preprocessing in Data Mining - page 260

Table 9.2 Enumeration of
data sets used in the
experimental study

Data set Data set
Abalone Appendicitis
Australian Autos
Balance Banana
Bands Bupa
Cleveland Contraceptive
Crx Dermatology
Ecoli Flare
Glass Haberman
Hayes Heart
Hepatitis Iris
Mammographic Movement
Newthyroid Pageblocks

Penbased Phoneme

Pima Saheart

Satimage Segment

Sonar Spambase

Specfheart Tae

Titanic Ve h i c l e

Vow e l Wine

Wisconsin Yeast

• C4.5 [ 92 ]: Awell-known decision tree, considered one of the top 10DMalgorithms

[ 120 ].

• DataSqueezer [ 28 ]: This learner belongs to the family of inductive rule extraction.

In spite of its relative simplicity, DataSqueezer is a very effective learner. The rules
generated by the algorithm are compact and comprehensible, but accuracy is to
some extent degraded in order to achieve this goal.

• KNN : One of the simplest and most effective methods based on similarities among

a set of objects. It is also considered one of the top 10 DM algorithms [ 120 ] and
it can handle nominal attributes using proper distance functions such as HVDM
[ 114 ]. It belongs to the lazy learning family [ 2 , 48 ].

• Naïve Bayes : This is another of the top 10 DM algorithms [ 120 ]. Its aim is to

construct a rule which will allow us to assign future objects to a class, assuming
independence of attributes when probabilities are established.

• PUBLIC [ 93 ]: It is an advanced decision tree that integrates the pruning phase

with the building stage of the tree in order to avoid the expansion of branches that
would be pruned afterwards.

• Ripper [ 32 ]: This is a widely used rule induction method based on a separate and

conquer strategy. It incorporates diverse mechanisms to avoid overfitting and to
handle numeric and nominal attributes simultaneously. The models obtained are
in the form of decision lists.