GEO-HYBRID: A Physics-Informed Python Framework for Liquefaction Potential Assessment and Integrated Geotechnical Interpretation

<title>Abstract</title> <p>https://github.com/houslast3/GEO-HYBRI Evaluating liquefaction risk and interpreting soil profiles traditionally rely on empirical relationships and classic geotechnical engineering formulations. Although recent machine learning-based models show high statistical performance on training datasets, the opaque nature of these algorithms and their susceptibility to overfitting limit their practical applicability and the reliability of their predictions in real engineering scenarios. This work presents GEO-HYBRID, a computational framework developed in Python that integrates established soil mechanics formulations with modern data-driven calibration techniques. Rather than replacing theoretical equations with black-box architectures, the methodology fully preserves fundamental physical laws—such as consolidation theory, shear strength, and effective stress calculation—using field data solely to calibrate parameters with direct physical meaning. The model was validated against multiple large-scale historical datasets, including real case histories of earthquake-induced liquefaction and cone penetration test (CPT) soundings. Results demonstrate that incorporating physical constraints and cyclic stress ratio corrections improves the predictive accuracy on unseen data, outperforming pure machine learning generalizations. Furthermore, numerical verification with continuous CPT profiles confirmed high convergence and computational stability relative to benchmark implementations in the literature. The framework establishes itself as a transparent, reproducible, and rigorous alternative for geotechnical interpretation and soil stability analysis under dynamic loading.</p>