Table 2. The results of ROC analysis in the framework of cross-validation.

AUC* and AUC** represent values obtained for one- and two-step optimization respectively. Significant P-values (using Bonferroni correction) and AUC values > 0.75 are marked in red and green, respectively. It can be seen that only alphascore and Shannon entropy-score based classifiers are effective in discriminating patients with epilepsy form controls under all studied conditions (one-step optimization, two-step optimization and one-step optimization restricted to normal and normal variant EEGs). Pdiagn and PnAED denote Kruskall Wallis P-values associated with the effect of diagnosis and nAED respectively and have been obtained for prefiltering and interval length values most commonly chosen in the process of classifier optimization (opt. band/interv. length). As can be seen the entropy-score classifiers are characterized by independence of AED-related effects.