Table 1.
Summary of the clinical data from the epileptic patients and the surgical outcome.
Table 2.
Set of predictive variables and their associated values. Cardinalities of each value are shown between brackets. Missing values are also indicated.
Figure 1.
Number of times that features were included in the different intermediate subsets selected by the race search feature selection for over 1,000 dataset resamplings.
Figure 2.
Estimated classification performance using LOOCV validation.
Only features available before surgery were included in this performance analysis. The x-axis reflects the size of the subset of features retained. A) The upper chart shows the estimated accuracy; whereas, B) the lower chart shows the associated area under the ROC curve. Note that the features for a given point on the x-axis can differ depending on the classifier used (see Table 3 for the respective feature subsets).
Table 3.
Variables ordered according to their frequency of selection during resampling and variable subset selection (occ: occurrence). Three different rankings are displayed - one for each of the classifiers used during the search.
Table 4.
Features and associated p-values obtained from a Wilcoxon signed-rank test comparing the values of each feature with the Engel output. in order of increasing p-value.
Table 5.
Probabilities of belonging to cluster 0 or cluster 1 for each case.