Skip to main content
. 2018 Dec 4;8:17593. doi: 10.1038/s41598-018-35934-y

Figure 4.

Figure 4

Error analysis of the predictions using ElemNet of a test set containing 25,662 compounds from our ten-fold cross validation. The left side shows that the predicted values are very close to the DFT-computed values. The right side illustrates the cumulative distribution function (CDF) of the prediction errors for ElemNet and Random Forest (the best performing conventional ML model) with elemental fractions (RF-Comp) and physical attributes (RF-Phys). Our error analysis demonstrates that the deep learning performs very well, achieving an MAE of 0.050 ± 0.000 eV/atom; predicting with an absolute error of less than 0.120 eV/atom for 90% of the compounds in our test set (right).