Table 3.

Model performance metrics

Metric	Across all participants	Average (SD) per participant
Volume of recorded data (days)	2070	90 (0)
Sample size (5 s clips)	14,904,000	648,000 (0)
# Falls confirmed by participants	37	1.6 (2.25)
# Stumbles confirmed by participants	31	1.3 (2.55)
# Falls-related events detected by the phone	72	3.3 (3.72)
True positives (correctly classified falls)	27	1.2 (2.1)
True negatives (correctly classified non-falls)	14,903,928	647999.6 (0.89)
False positives (wrongly classified non-falls)	45	1.95 (2.65)
False negatives (wrongly classified falls)	10	0.43 (0.84)
Daily false alarm rate	0.5	0.02 (0.03)
Accuracy	99.9997 %	99.9996 % (0.0004 %)
Sensitivity	73 %	68 % (39 %)*
Specificity	99.9998 %	99.9997 % (0.0004 %)
Precision	37.5 %	41 % (41 %)**
F1-score	0.495	0.51 (0.33)*

*n = 13, only participants who experienced at least one fall, were included; **n = 16, only participants whose phone classified at least one event as a fall, were included