Fig. 2: Performance of a bagging decision tree classifier for different sleep staging resolutions.
From: Automating sleep stage classification using wireless, wearable sensors
Confusion matrices (top), Receiver Operating Characteristic (ROC) curves (middle), and interquartile range (IQR) plots of model performance (bottom), obtained from leave-one-out cross-validation subject, for a two-stage wake vs. sleep classification, b three-stage wake vs. NREM vs. REM classification, c four-stage wake vs. light vs. deep vs. REM classification. ROC curves show the trade-off between sensitivity and specificity for a given model across subjects (line: mean; shading: standard deviation). Area under the ROC curve (AUROC) is listed for each stage; a value of 1.0 denotes a perfect classifier, whereas a value of 0.5 denotes a classifier that performs no better than random and has no predictive power. IQR plots illustrate how well the model generalizes across subjects, with smaller ranges indicating good performance and high generalizability irrespective of the subject (center line: median; box limits: upper and lower quartiles; whiskers: 1.5 × IQR; points: outliers).
