Skip to main content

Table 7 Top five important descriptors of LC selected by four DT-based ensemble learning models

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

RF

ExtraTrees

Selected descriptors

Feature importance

Selected descriptors

Feature importance

HeavyAtomCount

0.04649

NumRotatableBonds

0.03541

NumRotatableBonds

0.04381

HeavyAtomCount

0.03495

wing2_HeavyAtomCount

0.04329

wing1_NumRotatableBonds

0.02801

fr_unbrch_alkane

0.03315

wing2_HeavyAtomCount

0.02700

wing1_NumRotatableBonds

0.03218

wing1_HeavyAtomCount

0.02653

AdaBoost

GBM

Selected descriptors

Feature importance

Selected descriptors

Feature importance

HeavyAtomCount

0.08812

HeavyAtomCount

0.07017

NumRotatableBonds

0.06759

mesogen_HeavyAtomCount

0.05329

wing2_HeavyAtomCount

0.05722

NumRotatableBonds

0.04532

wing1_HeavyAtomCount

0.04705

mesogen_NumRotatableBonds

0.02774

mesogen HeavyAtomCount

0.04462

wing2_HeavyAtomCount

0.02718