Table 5.

Accelerometer-based cough detection results for the DNN classifiers. The values are averaged over 14 cross-validation folds. DNN classifiers have outperformed the shallow classifiers by a wide margin and a Resnet50 produces the highest AUC of 0.9888 in detecting cough events.

Classifier	ID	Best Feature	Best Classifier Hyperparameters	Performance
		Hyperparameters	(Optimised inside nested cross-validation)	Spec	Sens	Acc	AUC	${\sigma }_{\mathit{AUC}}$
CNN	C19	$\mathrm{\Psi }=16$, $C=5$	${\alpha }_1$=48, ${\alpha }_2$=2, ${\alpha }_3$=0.1, ${\alpha }_4$=32, ${\xi }_1$=128, ${\xi }_2$=210	83.47%	85.62%	84.55%	0.9243	0.002
	C20	$\mathrm{\Psi }=16$, $C=10$	${\alpha }_1$=24, ${\alpha }_2$=2, ${\alpha }_3$=0.3, ${\alpha }_4$=32, ${\xi }_1$=256, ${\xi }_2$=110	83.76%	87.56%	85.66%	0.9358	0.002
	C21	$\mathrm{\Psi }=32$, $C=5$	${\alpha }_1$=96, ${\alpha }_2$=2, ${\alpha }_3$=0.3, ${\alpha }_4$=32, ${\xi }_1$=128, ${\xi }_2$=150	76.98%	91.96%	84.47%	0.9272	0.002
	C22	$\mathrm{\Psi }=32$, $C=10$	${\alpha }_1$=48, ${\alpha }_2$=2, ${\alpha }_3$=0.3, ${\alpha }_4$=16, ${\xi }_1$=256, ${\xi }_2$=110	84.09%	86.41%	85.25%	0.9324	0.002
	C23	$\mathrm{\Psi }=64$, $C=5$	${\alpha }_1$=48, ${\alpha }_2$=2, ${\alpha }_3$=0.5, ${\alpha }_4$=32, ${\xi }_1$=256, ${\xi }_2$=230	85.47%	87.15%	86.31%	0.9339	0.002
	C24	$\mathrm{\Psi }=64$, $C=10$	${\alpha }_1$=96, ${\alpha }_2$=2, ${\alpha }_3$=0.3, ${\alpha }_4$=32, ${\xi }_1$=128, ${\xi }_2$=170	80.91%	90.73%	85.82%	0.9499	0.002
LSTM	C25	$\mathrm{\Psi }=16$, $C=5$	${\beta }_1$ = 128, ${\beta }_2$ = 0.0001, ${\alpha }_3$ = 0.3, ${\alpha }_4$ = 32, ${\xi }_1$ = 256, ${\xi }_2$ = 210	84.34%	90.82%	87.58%	0.9444	0.002
	C26	$\mathrm{\Psi }=16$, $C=10$	${\beta }_1$ = 128, ${\beta }_2$ = 0.01, ${\alpha }_3$ = 0.1, ${\alpha }_4$ = 32, ${\xi }_1$ = 128, ${\xi }_2$ = 110	85.37%	91.27%	88.32%	0.9504	0.002
	C27	$\mathrm{\Psi }=32$, $C=5$	${\beta }_1$ = 256, ${\beta }_2$ = 0.001, ${\alpha }_3$ = 0.3, ${\alpha }_4$ = 16, ${\xi }_1$ = 128, ${\xi }_2$ = 130	79.92%	94.31%	87.11%	0.9457	0.002
	C28	$\mathrm{\Psi }=32$, $C=10$	${\beta }_1$ = 128, ${\beta }_2$ = 0.001, ${\alpha }_3$ = 0.1, ${\alpha }_4$ = 32, ${\xi }_1$ = 256, ${\xi }_2$ = 150	86.41%	92.05%	89.21%	0.9572	0.002
	C29	$\mathrm{\Psi }=64$, $C=5$	${\beta }_1$ = 256, ${\beta }_2$ = 0.001, ${\alpha }_3$ = 0.3, ${\alpha }_4$ = 16, ${\xi }_1$ = 128, ${\xi }_2$ = 190	84.57%	92.79%	88.68%	0.954	0.002
	C30	$\mathrm{\Psi }=64$, $C=10$	${\beta }_1$ = 128, ${\beta }_2$ = 0.01, ${\alpha }_3$ = 0.5, ${\alpha }_4$ = 32, ${\xi }_1$ = 256, ${\xi }_2$ = 230	86.21%	89.13%	87.66%	0.9489	0.002
Resnet50	C31	$\mathrm{\Psi }=16$, $C=5$	Default Resnet50 (Table 1 in [73])	93.81%	97.09%	95.43%	0.9802	0.002
	C32	$\mathrm{\Psi }=16$, $C=10$	”	94.12%	98.58%	96.35%	0.9812	0.002
	C33	$\mathrm{\Psi }=32$, $C=5$	”	94.29%	98.79%	96.54%	0.9810	0.002
	C34	$\mathrm{\Psi }=32$, $C=10$	”	94.09%	99.33%	96.71%	0.9888	0.002
	C35	$\mathrm{\Psi }=64$, $C=5$	”	94.71%	98.23%	96.35%	0.9854	0.002
	C36	$\mathrm{\Psi }=64$, $C=10$	”	95.07%	97.89%	96.46%	0.9884	0.002