Skip to main content

Table 1 Statistical measures of the performance of the 10 machine learning algorithms and the median-based machine learning consensus predictor

From: Can human experts predict solubility better than computers?

 

RMSE

R2

ρ

NC

AAE

MLP

0.985

0.706

0.837

19

0.728

RF

1.165

0.583

0.736

20

0.802

Bagging

1.165

0.583

0.726

20

0.803

KNN

1.204

0.540

0.704

15

0.917

ExtraTrees

1.227

0.542

0.728

18

0.837

AdaBoost

1.235

0.545

0.708

19

0.851

PLS

1.265

0.507

0.670

15

0.980

SVM

1.280

0.520

0.694

16

0.925

SGD

1.429

0.577

0.752

11

1.185

Decision tree

1.813

0.260

0.530

17

1.198

ML median

1.140

0.601

0.762

18

0.778

  1. We assessed each machine learning method in terms of the root mean squared error (RMSE), coefficient of determination—which is the square of the Pearson correlation coefficient (R2), Spearman rank correlation coefficient (ρ), number of correct predictions within a margin of one log S unit (NC), and average absolute error (AAE)