Fig. 5: Feature Importance of XGBoost regression model.
From: XGBoost model for the quantitative assessment of stress corrosion cracking
a Global impact of attributes based on PI scores. The error bars represent the standard deviation of PI scores. Significant additions of Ni, Fe, Cr, and \({\sigma }_{R}\) collectively provide a weight to the model’s output from 0.573 to 0.661. b Local contributions via SHAP values. Here, data points are spread horizontally, reflecting the feature effect on the model’s predictions. A rightward shift suggests higher α values and increased SCC susceptibility, while a leftward shift implies lower α values and reduced SCC vulnerability. The SHAP values are expressed in the Box-Cox transformed scale of α. The colour gradient going from blue (low) to red (high) indicates the influence of the feature magnitude.
