Fig. 4: MSE of the GBR model versus number of descriptors. | npj Materials Degradation

Fig. 4: MSE of the GBR model versus number of descriptors.

From: Machine-learning-guided descriptor selection for predicting corrosion resistance in multi-principal element alloys

Fig. 4

Data are from numerous descriptor combinations from 1 to 13 for the absolute difference of predictions and experimental values, at least 19 out of 25 test data are within 2.5 times that of the experimental values. The lowest points represent the best regressor for a given number of descriptors. y-axis is limited for clarity to only those data that are <0.09. In actuality, there are 13Cn values for any given number of descriptors (n), as shown for n = 3 a total of 286. MSE does not show a trend with the number of descriptors and is minimum for a set of 5 descriptors (listed in figure).

Back to article page