Instance Selection Using Evolutionary Algorithms: An Experimental Study - Advanced Techniques in Knowledge Discovery and Data Mining - page 132

Training
set
selected

Data set

Instance
selection
algorithm

DM
algorithm

Model

Fig. 5.2. IS-TSS strategy.

Figure 5.2 shows a general framework for the application of an IS algorithm for
TSS. Starting from the data set D , the IS algorithm finds a suitable training set
selected, S .
In this work, we use EAs for IS-TSS following a stratified approach [27], [33],
which is outlined in Figure 5.3. The data set, D , is divided into two non-
overlapping sets with the 50% of the elements, T1 and T2 ( D = T1 T2 and T1
T2 = ), which are classically called strata . Then an IS algorithm may be applied
on them independently, obtaining two sets with the selected instances, S1 and S2 .
The final training set will be the union of these sets ( S = S1 S2 ).
This technique seems adequate for applying EAs as IS algorithms to DM
problems with large data sets (see Section 5.5.1).

Data set D

S 1

Training
set selected

T 1

S = S 1 S 2

Instance
selection
algorithm

T 2

S 2

Fig. 5.3. Stratified approach for IS-TSS.