Table 2.

Comparison of different modeling techniques for per-participant normalized VAS fatigue ratings

Model	No mean imp.			Mean imp.
	$R_{\%}^2$	MAE	RMSE	$R_{\%}^2$	MAE	RMSE
RF	26.1	0.69	0.86	24.0	0.71	0.89
BTE	26.0	0.69	0.86	23.6	0.71	0.89
SVM	25.8	0.68	0.85	24.6	0.70	0.88
NN	24.4	0.71	0.88	23.1	0.72	0.90
GAM	26.2	0.69	0.85	24.2	0.70	0.88
GLM	25.9	0.68	0.84	24.0	0.70	0.88

For all participants, we compare the performance of a GAM to a random forest (RF),²⁵ a boosted tree ensemble (BTE),²⁶ a linear support vector machine (SVM),²⁷ and a neural network (NN).²⁸ We evaluate all techniques when removing all data points with missing values (No mean imp.) and imputing each missing value by the mean feature value per participant (Mean imp.). Details in the STAR Methods section.