Fig. 3
Performance of the three feature classes (ROSA, BACD and G) in the prediction of the bandgap, EG, formation energy by atom, Ef, bulk modulus, KVHR, the vibrational entropy, S, the specific heat, CV and the effective dielectric constant, \({\varvec{\varepsilon}}\)eff. a The feature importance matrix for predicting the properties by the feature groups outlined in Section “Features”. b The receiver operating characteristic (ROC) for the metal/insulator classification model (described in the text). c–j The correlation plots for the prediction of the regression models for EG, Ef, KVHR, S, CV, \({\varvec{\varepsilon}}\)eff, the bandgaps for the materials in the QOMF database and the potential energy surfaces (PESs) for amorphized diamond unit cells, respectively