The input data for the feature selection included a total of 330 EEG functional connectivity variables consisting of 66 connectivity features for each of the 5 frequency bands. The model identified a total of 29 functional connectivity variables from multiple frequency bands connecting across the 12 default mode network seeds (Refer to Table 3 in the Methods Section and Figure S2 in Supplementary Materials). These connections included Delta—12 connections, Theta—6 connections, Alpha—4 connections, Beta—5 connections, and Gamma—2 connections. The 10-fold cross-validation for the λ1se threshold included all the 29 selected features, which were included in the subsequent implementation of the Random Forests classification model. The classification performance (to differentiate individuals with memory problems from those without) of the selected features as indicated by the area under the ROC curve (AUC) was 88.48%.
