Feature Evaluation by Filter, Wrapper, and Embedded Approaches - Feature Selection for Data and Pattern Recognition - page 38

3.6.2 Embedded DRSA Ranking

The second batch of performed tests employed an embedded DRSA ranking (see
Table 3.1 ) in backward reduction of features, again for both connectionist and
rule-based classifiers. Classification results for artificial neural networks are pre-
sented in Fig. 3.3 , with reduction of the highest ranking features in Fig. 3.3 a and
elimination of lowest ranking variables in Fig. 3.3 b. The pattern visible in the two
graphs shows close resemblance to the one observed for the first tested ranking. We
can also note that in most cases there are bigger differences between the maximal
and minimal performances of the tested networks.
Figure 3.4 illustrates reduction of characteristic features with highest and lowest
rank, based on the embedded DRSA approach, for decision algorithms. For the
decreasing order in the initial phase of reduction the results are acceptable, but once

(a) (b)

Fig. 3.3 ANN classification accuracy in relation to the number of considered features, observed in
sequential backward elimination process while employing feature ranking obtained by an embedded
approach based on relative reducts: a decreasing order, b increasing order. For each median there
is indicated maximal and minimal performance

Fig. 3.4 DRSA classification accuracy in relation to the number of considered features, observed
in sequential backward elimination process of all rules on examples algorithm, while employing
feature ranking obtained by embedded approach based on relative reducts, for a decreasing order
(Most series), and increasing order (Least series)