Table 2. Predictive performance of random forest models.

Model diagnostics describing overall performance when applied to predict host category of held-out coronaviruses not used for model training. CI denotes confidence interval, Kappa denotes Cohen’s Kappa statistic, mAUC denotes multiclass area-under-curve statistic, and F1macro denotes F1 score calculated using macro-averaging (performance on each host category weighted equally).

Predictor features	Accuracy (95% CI)	Kappa	mAUC	F1macro
Spike protein	0.735 (0.700, 0.769)	0.696	0.898	0.757
Whole genome	0.728 (0.687, 0.766)	0.688	0.902	0.758