Fig. 6: ML performance as a function of the number of top-ranking features. | npj Computational Materials

Fig. 6: ML performance as a function of the number of top-ranking features.

From: Coupling physics in machine learning to predict properties of high-temperatures alloys

Fig. 6

Accuracy of five trained ML models (BR Bayesian ridge regression; LR linear regression; NN nearest neighbor; RF random forest; SVM support vector machines regression) in predicting yield strength. These models were trained with synthetic features populated from high-throughput calculation as a function of the number of top-ranking features in the truncated (≤650°C) dataset. The hyperparameters of each model were tuned up to 150 iterations to obtain optimum parameters. Each model was trained ten times to determine the average accuracy and its standard deviation (error bar).

Back to article page