Table 6.

Mean values ± standard deviation (in percent) of the performance measures (F2 and F1 scores, and AUC) for the overnight prediction (10 pm to 7 am) given for the baseline models logistic regression (LR) and random forest (RF) and the Deep Neural Network (DNN) models.

			Time series characteristics
Data sets		Glucose literature features	Glucose	Glucose & static	Glucose & physiological	Glucose, physiological & static
LR	F2	56.4 ± 20.9	50.6 ± 24.0	52.1 ± 19.8	56.3 ± 27.8	63.0 ± 23.7
	F1	56.0 ± 12.0	42.7 ± 18.6	43.6 ± 15.4	49.3 ± 24.2	54.8 ± 20.3
	AUC	68.3 ± 13.1	59.4 ± 15.4	58.9 ± 14.0	66.0 ± 17.2	69.0 ± 16.2
RF	F2	40.8 ± 23.1	64.4 ± 26.0	56.0 ± 26.3	61.3 ± 30.3	56.6 ± 29.6
	F1	47.3 ± 25.2	58.3 ± 22.6	48.3 ± 22.3	60.9 ± 27.6	55.4 ± 25.9
	AUC	66.4 ± 11.9	73.5 ± 14.5	65.3 ± 15.3	75.2 ± 16.6	72.6 ± 14.9
DNN	F2	n.a.	57.5 ± 19.1	59.5 ± 12.9	62.0 ± 11.3	49.1 ± 22.3
	F1	n.a.	41.9 ± 14.1	40.4 ± 7.5	42.1 ± 6.9	37.6 ± 16.6
	AUC	n.a.	53.9 ± 13.7	48.2 ± 8.6	51.0 ± 7.6	51.7 ± 14.0

The standard deviations are calculated as average values of the five runs. For LR and RF, different features were calculated (Literature Features or Time Series Characteristics). The used machine learning pipeline is illustrated in Figure 1. AUC, Area Under the Receiver Operating Characteristic Curve; n.a., not applicable. Best F2 scores per data set are marked in bold.