A Machine Learning-Driven Geophysical–Geotechnical Approach for Improved Engineering Site Assessment

Subsurface geological formations are vital for validating deep engineering design assumptions, particularly in weathered terrains where unstable ground conditions pose risks. Geophysical investigations often face challenges due to inverse problem uncertainties and inadequate subsurface data. While resistivity and seismic P-wave velocity (Vp) imaging offer valuable insights, subsurface complexity necessitates integrated approaches for reliable characterization. This study introduces a machine learning-assisted geophysical–geotechnical framework combining electrical resistivity tomography, seismic refraction tomography, and borehole-based standard penetration tests (SPT-N). ML optimization metrics, including k-means clustering, PCA, Silhouette, elbow, and supervised linear regression, enhance analytical precision. A field survey over an 800 m segment in Kabota-Tawau, Sabah, Malaysia, utilized 490 collocated resistivity–Vp datasets to optimize cluster identification and interpretation accuracy. The analysis delineated four lithological units based on resistivity and Vp variations correlating with surface-subsurface properties. Clustering demonstrated strong performance, with an R² value approaching 1, a Silhouette score of 0.78, and an 88% reduction in the sum of square errors. Vulnerable zones, including weathered layers, fractures, and faults, were identified as critical for geotechnical consideration. In contrast, relatively weathered bedrock with hard-to-very-hard properties was deemed optimal for deep structural foundations requiring minimal reinforcement. This non-invasive approach enhances subsurface characterization, offering a reliable framework for construction site suitability and groundwater resource identification. It provides significant value for sedimentary regions with similar geological settings, advancing geotechnical and environmental planning.