Defining New Statistical Features for Geotechnical Properties: Exploring Higher-Order Dependencies in Mixed Domain Spaces with the Minimum Information Dependence Model

Understanding the complex interdependencies among soil parameters is crucial for designing and maintaining civil engineering structures. Traditional modeling often assumes linearity and normality, potentially overlooking the nonlinear and higher-order relationships inherent in geotechnical data. This study evaluates the effectiveness of Minimum Information Dependence Modeling (MIDM), a novel joint probability distribution model, in analyzing dependencies within heterogeneous geotechnical datasets comprising both quantitative and qualitative variables. Applying MIDM to the real soil database, we computed conditional scores using a pseudolikelihood estimator based on Besag's approach to estimate strata labels. The results demonstrate that MIDM successfully captures complex non-linear relationships, providing a more nuanced understanding of subsurface conditions and enhancing anomaly detection and model validation in geotechnical risk assessments. Despite the high computational costs associated with the sampling-based approach, MIDM shows promise for transforming traditional practices toward more data-driven and quantitative methodologies. Future research will focus on optimizing computational efficiency and integrating MIDM with other data-driven techniques to construct high-density three-dimensional ground models, aiming to establish MIDM as a reliable tool in geotechnical investigations.

Keywords: Minimum information dependence model, Mixed-domain, Big indirect database, Data-centric geotechnics, Patterns extraction, 3D subsurface geological model, Stratigraphy.