Predicting cognitive scores from wearable-based digital physiological features using machine learning: data from a clinical trial in mild cognitive impairment

Table 5 Performance of the best models predicting intra-individual changes in NTB scores

Cross-validation method	Cognitive measure	Machine Learning algorithm	Number of used features	Feature selection p-value	r	r gain	rho	rho gain	MAE
Subject-based	Global Cognition	Random Forest	19	0.05	0.11	0.11	0.12	0.12	0.11
	Executive Function	Random Forest	15	0.001	0.61	0.61	0.44	0.44	0.07
	Processing Speed	ElasticNet	3	0.001	0.20	0.20	0.23	0.23	0.14
	Memory Immediate	ElasticNet	7	0.05	0.29	0.35	0.31	0.41	0.21
	Memory Delayed	Random Forest	6	0.05	− 0.11	− 0.11	− 0.05	− 0.05	0.19
Interval-based	Global Cognition	XGBoost	4	0.01	0.20	0.20	0.06	0.06	0.09
	Executive Function	Random Forest	15	0.001	0.77	0.77	0.48	0.48	0.06
	Processing Speed	XGBoost	3	0.001	0.45	0.45	0.30	0.30	0.14
	Memory Immediate	XGBoost	3	0.01	0.21	0.54	0.22	0.20	0.25
	Memory Delayed	XGBoost	6	0.05	0.06	0.06	− 0.01	− 0.01	0.17

ISSN: 1741-7015