Table 1.

Hyperparameter tuning of baseline kNN and RF classifier models. The parameters of the optimal models obtained in the nested grid search are shown in bold.

Model	Hyperparameters (optimal are shown in bold)	Cross- validation F1 (macro) score	External (hold- out) set F1 (macro) score
kNN	fingerprint radius: [2, 3, 4, 5] fingerprint length: [1536, 2048, 2560] number of nearest neighbors: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10] weights: [‘uniform’, ‘distance’]	0.784	0.796
RF	fingerprint radius: [2, 3, 4, 5] fingerprint length: [1536, 2048, 2560] n_estimators: [50, 100, 150, 200, 250, 300] max_features: [0.001, 0.002, 0.005, 0.01, 0.02,.., 1.] min_samples_split: [2, 3, 4]	0.853	0.861