Table 5.

Evaluation Metrics for Different Models (Results from Ablation Study)

	Cohen’s Kappa	Macro F1	Mean Per-Class Sensitivity	Accuracy
One-to-one	0.305	0.461	0.538	0.493
20-to-20	0.489	0.614	0.640	0.665
One-step 40-to-20	0.476	0.612	0.670	0.643
SoundSleepNet	0.532	0.644	0.655	0.703

Notes: Bolded values indicate the highest score. The model was trained with the PSG audio training set and evaluated with the PSG audio test set.