Fig. 3: Workflow and performance metrics of fingerprint-like descriptor-based logistic regression (FLDLR) of the configuration stability. | Nature Communications

Fig. 3: Workflow and performance metrics of fingerprint-like descriptor-based logistic regression (FLDLR) of the configuration stability.

From: Automated exploitation of the big configuration space of large adsorbates on transition metals reveals chemistry feasibility

Fig. 3

a The subgraph space is decomposed using the method described in ref. 33. b–e shows the recall, precision, F1 score, and selectivity of the resulting model for 11 surfaces, respectively. The inset indicates the metric equation where T, F, P, and N indicate true, false, positive, and negative, respectively. The threshold is a tunable probability decision boundary to predict stability, which results in different performances for the metrics considered in this study. The light and dark shaded region indicates 99 and 95% confidence interval, respectively, and the gray and black lines indicate values for 11 surfaces and their mean. The model is trained with adsorbates with ≤2 C and O atoms and tested on adsorbates with 3 C and O atoms.

Back to article page