Table 3.

Performance comparison on the OSASUD dataset using different positional encodings of the Transformer architectures.

Architecture	AUC	Accuracy	F1	Sensitivity	Specificity	Precision
Transformer encoder only	0.8738	0.7898	0.8092	0.7473	0.8526	0.8822
Transformer encoder with naïve position embeddings	0.8788	0.7969	0.8105	0.7665	0.8366	0.8598
Transformer encoder with sinusoidal positional embedding	0.8799	0.7983	0.8118	0.7678	0.8381	0.8610
Transformer encoder with Learned Positional Embedding	0.8890	0.7995	0.7931	0.8285	0.7745	0.7605