Table 2. Prediction Accuracies (R²) for the Test1–Test4 Datasets of the BHC Reaction^a.

	Test1	Test2	Test3	Test4
one-hot-RF	0.69 (0.00)	0.67 (0.00)	0.50 (0.00)	0.48 (0.00)
random-RF	0.69 (0.00)	0.82 (0.00)	0.52 (0.00)	0.42 (0.00)
yield-BERT	0.84 (0.01)	0.83 (0.01)	0.74 (0.01)	0.49 (0.02)
T5Chem	0.82 (0.01)	0.91 (0.01)	0.76 (0.01)	0.55 (0.01)
XGBoost	0.88 (0.00)	0.89 (0.00)	0.60 (0.00)	0.58 (0.00)
MPNN-transformer	0.87 (0.01)	0.88 (0.01)	0.59 (0.03)	0.64 (0.01)
MPNN-transformer(3.0m)	0.87 (0.01)	0.90 (0.00)	0.74 (0.01)	0.61 (0.01)
*DFT-RF(2)	0.80	0.77	0.64	0.54
*Mol2Vec-MPNN(9)	0.92	0.88	0.60	0.39

^aFor each model and dataset, the average (standard deviation) of the R² value for the Test1–Test4 datasets of the BHC reaction using the five ensemble models is reported. *For DFT-RF(2) and Mol2Vec-MPNN,⁹ the reported values are given, so direct comparison may not be appropriate. For each test dataset, the three highest R² values are highlighted in bold.