Table 3.

Performance of different regressor models we tested.

Model	MAE	MSE	Accuracy (%)	Kendall’s τ	MAPE (%)	PCC	Spearman’s ρ
SVR	0.5861	0.5586	51.94	0.5044	32.14	0.6388	0.6329
Random forest regressor	0.5920	0.5482	49.28	0.5116	33.5	0.6518	0.6389
AdaBoost regressor	0.5926	0.5499	45.6	0.5038	33.75	0.6219	0.6317
XGBoost regressor	0.5904	0.5553	51.33	0.4989	32.57	0.6417	0.6282
LightGBM regressor	0.5802	0.5364	50.92	0.5147	32.01	0.6563	0.6429
Shallow neural network-I (one trainable layer)	0.6154	0.6007	46.83	0.4810	33.9	0.6097	0.6100
Shallow neural network-II (two trainable layers)	0.6069	0.6162	51.33	0.4813	32.42	0.6004	0.6044

Performance was reported based on leave-one-patient-out cross-validation. For each performance metric, the best value is highlighted in bold text. Some of the metrics are abbreviated for the simplicity of presentation.

MAE mean absolute error (points), MSE mean squared error (points), MAPE mean absolute percentage error (%), PCC Pearson’s correlation coefficient.