Fig 1.
Gradient boosting decision tree algorithm.
Fig 2.
Random forest algorithm.
Table 1.
Definitions of the features selected [50].
Fig 3.
Prediction performance of four ML algorithms (GBDT, RF, SVR and MLP) without the feature Gap_nosoc.
(a) Error levels of ML predicted bandgaps (b) Fitness between predicted bandgaps and true bandgaps on the training sets (c) Fitness between predicted bandgaps and true bandgaps on the test sets.
Fig 4.
Prediction performance of four ML algorithms (GBDT, RF, SVR and MLP) with the feature Gap_nosoc.
(a) Error levels of ML predicted bandgaps (b) Fitness between predicted bandgaps and true bandgaps on the training sets (c) Fitness between predicted bandgaps and true bandgaps on the test sets.
Table 2.
Statistics of predicted bandgaps by SVR, GBDT, RF and MLP algorithms based on an 8-dimensional feature space.
Table 3.
Statistics of predicted bandgaps by SVR, GBDT, RF and MLP algorithms based on a 9-dimensional feature space.
Fig 5.
Feature importance evaluation.
(a) 8-dimensional feature space (b) 9-dimensional feature space.