Fig. 3: Sleep intrusions and wake instability predict sleep groups.
From: A continuous approach to explain insomnia and subjective-objective sleep discrepancy
A Classifier average performance as a function of the number of features included as predictors for binary (black, only SOSD- and SOSD+ ) and 3 class problems (red, GS, SOSD- and SOSD+ ) using size-balanced subsamples of data. For the 3-class problem all random subsamples of SOSD− and SOSD+ comprised 104 subjects, while for the binary problem random SOSD− subsamples of 199 subjects were taken, so all the sleep groups had the same sample size. Solid line and shaded area correspond to the average and 1 standard deviation, respectively, of 60 iterations of a repeated 5-fold cross-validated classifier, using 20 random subsamples each time. B, C Average confusion matrix for the 3 and 2-class problem, corresponding to the vertical lines in (A). D, E 2D Histograms of the class prediction accuracy for each subject in the binary problem and sleep quality metrics for SOSD− and SOSD+ , respectively. Color scale represents the percentage of the population and vertical dashed line represents the cut-off for the diagnosis associated with each sleep metric.
