Table 1 Detailed comparison of 13 different featurization schemes for prediction of explosive energy with kernel ridge regression, ranked by MAEtest. These variance in MAEs between folds was less than 0.01 in all cases. Hyperparameter optimization was used throughout with nested 5-fold cross validation. The metrics are averaged over 20 train-test sets using shuffle split with 80/20 splitting.

From: Applying machine learning techniques to predict the properties of energetic materials

name MAEtrain MAEtest MAPEtest \({{\boldsymbol{R}}}_{{\bf{train}}}^{{\bf{2}}}\) \({{\boldsymbol{R}}}_{{\bf{test}}}^{{\bf{2}}}\) rtrain rtest
E-state + CDS + SoB 0.244 0.334 8.93 0.88 0.76 0.88 0.79
CDS + SoB 0.247 0.335 9.32 0.88 0.75 0.88 0.79
E-state + custom descriptor set 0.224 0.345 9.50 0.89 0.75 0.90 0.79
SoB + OB100 0.256 0.358 10.50 0.87 0.61 0.87 0.70
sum over bonds (SoB) 0.280 0.379 10.69 0.84 0.67 0.84 0.71
truncated E-state 0.260 0.414 12.65 0.85 0.66 0.85 0.70
custom descriptor set (CDS) 0.398 0.432 12.92 0.68 0.57 0.68 0.63
Bag of Bonds (BoB) 0.213 0.467 12.60 0.89 0.54 0.90 0.60
Oxygen balance1600 0.419 0.489 15.66 0.67 0.41 0.68 0.56
Summed Bag of Bonds 0.262 0.493 13.63 0.85 0.18 0.85 0.56
Coulomb matrix eigenvalues 0.314 0.536 15.73 0.81 0.37 0.82 0.48
Oxygen balance100 0.444 0.543 17.46 0.59 0.44 0.62 0.57
Coulomb matrices as vec 0.395 0.672 21.86 0.57 0.05 0.67 0.20