Irrelevant Feature and Rule Removal for Structural Associative Classification Using Structure-Preserving Flat Representation - Feature Selection for Data and Pattern Recognition - page 199

T 0 T 1 T 2 T 3

'a' 'b' 'b' 'l'

'm' 'n'

'b' 'c' 'c' 'b' 'd'

'd'

'e' 'e'

'n'

'e'

T 4 T 5 T 6

'k' 'b' 'a'

'l' 'n' 'a' 'd' 'b' 'f'

'm'

'g'

'c'

'c'

'd' 'h'

'e' 'i'

'f'

Fig. 10.2 Example of a tree-structured database T db consisting of 7 transactions

Table 10.1 Example of tree
transactions

Tree database ( T db ) Pre-order string encoding

− − − − −

T 0 'a b d n 1e 1 1 1c 1'

T 1 'b c − 1be − 1 − 1'

T 2 'b d e − 1 − 1'

− −

T 3 'l m 1n 1'

T 4 'k l m − 1 − 1n − 1'

T 5 'bacf − 1 − 1 − 1d − 1'

T 6 'a b c d − 1e − 1 − 1 − 1fgh − 1i

− 1 − 1 − 1'

reducing the number of rules that need to be generated from the association rule
mining algorithm, while closely maintaining the integrity of the original data [ 18 ].
Additionally, rules described with fewer attributes are also expected to perform better
when classifying future cases; hence, they will have better generalization power than
do the more specific rules that take many attributes into account. Besides, the patterns
extracted will also be simpler and easier to analyse and understand. Determining the
relevant and irrelevant attributes poses a great challenge to many data mining algo-
rithms [ 36 ].