Fig. 2: Precision and recall. | npj Computational Materials

Fig. 2: Precision and recall.

From: From individual elements to macroscopic materials: in search of new superconductors via machine learning

Fig. 2: Precision and recall.

a The performance of the classifier are challenged in the plane precision vs. recall. Green symbols refer to the SuperCon database, potentiated with data from COD. Blue symbols refer to the Hosono database. Each symbol is computed as the average over 60 independent realizations and the shadowed region reflects the associated standard deviation. b the precision vs. recall curve is plotted for the Hosono database by using the majority rule, as described in the main text: a material is classified superconducting if the majority of runs (over 60 realizations), each associated to ϵth = 0.85, leads to this conclusion. The horizontal dashed line is a benchmark reference computed by assuming random guessing for the Hosono dataset.

Back to article page