Table 1.

Performance metrics and 95% confidence intervals (CIs) of the gradient-boosted decision trees model (Extreme Gradient Boosting) with the top 68 features, the Juniper fall risk assessment score, and other machine learning models (logistic regression and multilayered perceptron) for the 3-month prediction of fall.

Variable	Extreme Gradient Boosting	Logistic regression	Multilayered perceptron	Juniper fall risk
Area under the receiver operating characteristic curve (95% CI)	0.846 (0.794-0.894)	0.711 (0.645-0.773)	0.697 (0.624-0.765)	0.621 (0.547-0.693)
Sensitivity (95% CI)	0.706 (0.577-0.833)	0.706 (0.553-0.859)	0.706 (0.571-0.833)	0.351 (0.217-0.485)
Specificity (95% CI)	0.848 (0.809-0.888)	0.614 (0.560-0.668)	0.612 (0.566-0.657)	0.883 (0.854-0.911)
Positive likelihood ratio	4.647	1.828	1.813	3.014
Negative likelihood ratio	0.346	0.479	0.481	0.733
Diagnostic odds ratio (95% CI)	13.400 (6.026-29.796)	3.816 (1.764-8.256)	3.766 (1.741-8.147)	4.113 (1.881-8.995)
True positive	24	24	24	12
True negative	268	194	193	279
False positive	48	122	123	37
False negative	10	10	10	22
F1a	0.393	0.262	0.248	0.289

^aF score is defined as the harmonic mean between precision and recall.