Table 2.

Mean squared error (MSE) of the best three-feature and four-feature combinations of the linear regression, support vector, k-nearest neighbors, and random forest regression models for predicting aggregation rates. Hyperparameters are set to the default Scikit-learn parameters.

Regression Models	Three-feature	MSE	Four-feature	MSE
Linear	SAP_pos_CDRH3 SCM_pos_CDRL3 SCM_neg_CDRH3	0.433	SAP_pos_CDRH3 SCM_pos_CDRL1 SCM_pos_CDR SCM_pos_Hv	0.457
Nearest neighbors (neighbor numbers = 5)	SAP_pos_CDRL3 SCM_pos_CDRH3 SCM_neg_Fv	0.366	SCM_pos_CDRH3 SCM_neg_CDRH2 SCM_neg_Hv SCM_neg_Fv	0.319
Random forest (max_depth = None)	SAP_pos_Hv SCM_pos_CDRH3 SCM_neg_Fv	0.367	SAP_pos_Hv SCM_pos_CDRH3 SCM_pos_Lv SCM_neg_Fv	0.364
Support vector (C = 1.0, ε = 0.1)	SCM_pos_CDRH3 SCM_neg_CDRH2 SCM_neg_Fv	0.307	SCM_pos_CDRH3 SCM_neg_CDRH2 SCM_neg_CDRL2 SCM_neg_Fv	0.301