fig4

Figure 4. Further feature selection for the ML models. (A) Bar chart of feature importance and ring chart of feature contribution ratios; (B) Summary plot of SHAP values; The optimal subset selection for the (C) RT and (D) HT dataset. ML: Machine learning; SHAP: SHapley Additive exPlanations; RT: room-temperature; HT: high-temperature; σ: stress amplitude; TS: tensile strength; YS: 0.2% proof stress; dA, dB, dC: non-metallic inclusions; EL: elongation; TFS: true fracture stress; RA: reduction of area; Is: ingot size; Ags: austenite grain size number; Rr: reduction ratio; R²: the coefficient of determination; RMSE: root mean square error.