Figure 9.

Results of the ML model. a) Fitting results of calculated binding energies and predicted binding energies of CO*. R², MAE and RMSE are computed to estimate prediction errors and total training time for one thousand trials. The inset shows the convergence of GBR model accuracy for five cross‐validation splits of the data. b) Feature‐importance scores of descriptors for the GBR prediction of binding energies of CO* on various alloy surfaces. c–e) Performance of the ML model for predicting CO* binding energies on fcc(111) of NSAs and binding energies of various intermediates on SAAs surfaces. Insets in the lower right corner show the prediction error distribution of the GBR model. Reproduced with permission.^{[ 127 ]} Copyright 2020, The Royal Society of Chemistry.