Skip to main content

Advertisement

Table 6 RF prediction of DM with subsets of descriptors from models C and F

From: Machine learning for the prediction of molecular dipole moments obtained by density functional theory

Model/no descriptors MAE (D) R2/RMSE (D)
Training set
 C/75a 0.515 0.826/0.769
 C/100a 0.517 0.826/0.771
 C/125a 0.518 0.826/0.7709
 C/32b 0.521 0.824/0.773
 C/39c 0.523 0.824/0.775
 C/296d 0.529 0.821/0.782
 F/75a 0.482 0.844/0.731
 F/100a 0.482 0.844/0.731
 F/125a 0.483 0.844/0.732
 F/34b 0.501 0.838/0.745
 F/41c 0.499 0.838/0.743
 F/297d 0.503 0.832/0.755
Test set
 C/75a 0.502 0.847/0.744
 C/100a 0.506 0.846/0.748
 C/125a 0.505 0.845/0.748
 C/32b 0.509 0.845/0.747
 C/39c 0.512 0.844/0.749
 C/296d 0.519 0.843/0.758
 F/75a 0.468 0.864/0.704
 F/100a 0.466 0.865/0.702
 F/125a 0.466 0.867/0.699
 F/34b 0.481 0.860/0.713
 F/41c 0.479 0.860/0.710
 F/297d 0.485 0.852/0.725
  1. aOOB estimation for the training set
  2. bUsing the mean decrease in accuracy measure of importance for the descriptors in the RF algorithm
  3. cUsing the the CFS with BestFirst routine from Weka
  4. dUsing the the CFS with GreedyStepwise routine from Weka
  5. eUsing the the CFS with PSOSearch routine from Weka