A SHAP value method for ultimate strength prediction of stiffened panel: A data-driven tool in engineering

Artikel i vetenskaplig tidskrift, 2026

This study introduces a technique to derive the primary variables of the empirical formula for predicting the ultimate strength of stiffened panels by utilising the SHAP (SHapley Additive exPlanations) value, which is a means to explain the output of a machine learning model. Existing empirical equations, which can predict the ultimate compressive strength of stiffened panels, adopt plate- and column-slenderness ratios as the governing variables that may not suffice to capture the nonlinearity of ultimate limit state (ULS) behaviour. Recent studies have enhanced the accuracy by adopting additional dimensionless variables (hw/tw; Ipz/Isz), but there is still a lack of clear method for determining the relative importance of these variables. The present study, therefore, proposes a new procedure and criterion for defining the variables required for empirical expression development based on the feature importance of variables using SHAP values. The proposed systematic procedure may enable the define a configurable empirical expression form, and the extracted variables are substituted for each expression. The polynomial fitting is performed using a Pseudo Inverse Matrix to verify the improvement in accuracy compared to the existing empirical expression. The outcomes of this study, i.e., the applicability of the SHAP value method to select variables and the accuracy of the ULS prediction results, may be a reliable resource for predicting the ultimate compressive strength of local structures used in ships and offshore structures.