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

Figure 3.

Figure 3

Impact of training dataset size on the prediction accuracy of ElemNet (DNN model) using elemental compositions only and the best conventional ML model, Random Forest, with either raw elemental compositions (RF-Comp) and physical attributes (RF-Phys). The training and test sets are created during the ten-fold cross validation from the OQMD; different random subsets of the training set with sizes ranging from 464 to 230, 960 are created using a logarithmic spacing for this analysis. Training dataset size has more impact on ElemNet (deep learning model) compared to Random Forest models, but ElemNet performs better than Random Forest for all size greater than 4k.