GMDH-Derived Correlation for investigating Soil Internal Friction Angle from SPT and Physical Indices

This study develops and evaluates a Group Method of Data Handling (GMDH)–based polynomial model to estimate the internal friction angle (ϕ) of soils from in-situ measurements. The modeling framework uses ۳۱۸ data sets compiled from multiple geographic regions, thereby capturing broad variability in depositional history and soil fabric. Predictor variables comprise the Standard Penetration Test (SPT) blow count (N) together with seven influential in-situ/geotechnical parameters: effective overburden stress, fines content, Atterberg limits, water content, dry unit weight, and particle-size descriptors, including the percentage passing the No. ۴ sieve. The GMDH algorithm automatically selects structure and coefficients, yielding an interpretable polynomial network that balances accuracy and parsimony. Model performance was assessed using standard statistical indicators and external verification procedures, and input significance was examined through sensitivity analysis to quantify the relative contribution of each parameter. Results demonstrate that the proposed model attains high predictive accuracy for ϕ across the heterogeneous, globally sourced database, underscoring its robustness to site-specific biases. Given its reliance on commonly available field and laboratory indices, the approach offers a practical and powerful tool for preliminary design and data-driven characterization of shear strength parameters where direct shear or triaxial testing is limited. Overall, the findings support GMDH as an effective methodology for identifying reliable empirical relationships between ϕ and routinely measured soil properties.

کلیدواژه ها:

GMDH ، Internal friction angle ، Sensitivity analysis ، SPT ، Atterberg ، correlation