Skip to main content

Table 9 Top five important descriptors selected by three blending methods

From: Comparison and improvement of the predictability and interpretability with ensemble learning models in QSPR applications

Uniform blending

Linear blending

Selected descriptors

Feature importance

Selected descriptors

Feature importance

HeavyAtomCount

0.05993

HeavyAtomCount

0.04781

NumRotatableBonds

0.04803

NumRotatableBonds

0.04219

wing2_HeavyAtomCount

0.03827

wing2_HeavyAtomCount

0.03471

mesogen_HeavyAtomCount

0.03484

wing1_HeavyAtomCount

0.03006

wing1_HeavyAtomCount

0.03290

fr_unbrch_alkane

0.02897

Any blending

  

Selected descriptors

Feature importance

  

HeavyAtomCount

0.06520

  

NumRotatableBonds

0.05236

  

wing2_HeavyAtomCount

0.04321

  

wing1_HeavyAtomCount

0.03636

  

mesogen_HeavyAtomCount

0.03556