Table 3.

Performance metrics of all models used to predict food group consumption and daily diet quality. This table presents the performance of four machine learning models—Stochastic gradient boosted decision trees (SGBDT), random forests (RF), hurdle SGBDT, and hurdle RF—applied to predict intake (serves) of individual food groups and daily diet quality scores. Outcomes include fruit, vegetable, grain, meat and alternatives, dairy and alternatives, discretionary foods, and an overall daily diet quality measure. For each outcome, the model with the lowest MAE (best performance) is marked with an asterisk (*) with other metrics bolded. Lower RMSE and MAE, and higher R² values indicate better model performance.

		Performance metrics
Outcome	Model	RMSE1	MAE2	R squared
Fruits	SGBDT	0.63	0.35	0.049
	RF	0.56	0.30*	0.418
	Hurdle SGBDT	0.71	0.59	0.023
	Hurdle RF	0.69	0.44	0.598
Vegetables	SGBDT	1.57	0.94	0.099
	RF	1.33	0.75*	0.463
	Hurdle SGBDT	1.12	0.99	0.031
	Hurdle RF	1.54	1.00	0.517
Grains	SGBDT	1.29	0.96	0.096
	RF	1.03	0.72	0.496
	Hurdle SGBDT	0.94	0.85	0.021
	Hurdle RF	0.82	0.55*	0.616
Meat and alternatives	SGBDT	0.80	0.52	0.097
	RF	0.72	0.45	0.439
	Hurdle SGBDT	0.69	0.59	0.022
	Hurdle RF	0.62	0.40*	0.632
Dairy and alternatives	SGBDT	0.49	0.33	0.055
	RF	0.43	0.28*	0.441
	Hurdle SGBDT	0.57	0.50	0.017
	Hurdle RF	0.44	0.30	0.562
Discretionary foods	SGBDT	1.38	0.79	0.129
	RF	1.12	0.59*	0.552
	Hurdle SGBDT	0.83	0.68	0.027
	Hurdle RF	1.23	0.72	0.620
Daily diet quality	SGBDT	14.57	11.96	0.113
	RF	14.41	11.86	0.119

*Best performing. Bold – best performing metrics

¹RMSE - Root Mean Squared Error

²MAE - Mean Absolute Error