Table 2.

Window-level classification performance of the 48D and 15D feature sets on WESAD.

Feature set	Setting	Acc.	F1	ROC-AUC	Prec.	Spec.
48D	Baseline RF	0.837	0.639	0.885	–	–
15D	Baseline RF	0.822	0.606	0.874	–	–
48D	Cross-fit + sigmoid calib + τ*	0.866	0.733	0.885	0.941	0.984
15D	Cross-fit + sigmoid calib + τ*	0.848	0.685	0.874	0.935	0.984

τ^* denotes the validation-selected optimal decision threshold.