Global Clay Preconsolidation Stress Analysis

Abstract

This Python software reproduces the data screening, model development, calibration, hyperparameter tuning, validation, interpretability analysis, and performance assessment presented in the accompanying manuscript, “Geotechnically Informed Prediction of Preconsolidation Stress from Clay Index Properties: Recalibration, Nonlinear Learning, and Stress-History Interpretation.” The program analyzes the Global Database of Effective Preconsolidation Stress and Index Properties of Clays and evaluates empirical, statistical, machine-learning, and physics-guided methods for estimating effective preconsolidation stress from clay index properties and current effective vertical stress. The implemented methods include the original and recalibrated Kootahi–Mayne framework, Ridge regression, support-vector regression, random forest, extremely randomized trees, gradient boosting, XGBoost, LightGBM, multilayer perceptron, physics-guided XGBoost, and data-driven OCR threshold and multiregime analyses. The software uses a fixed random state of 42 and exports screening records, fitted parameters, model-performance metrics, cross-validation results, test-set predictions, feature-importance results, analysis figures, and fitted model files. Execution instructions and software requirements are provided in the accompanying README and requirements files. The required input database is publicly available at Global Clay Preconsolidation Stress Database Version 1.0.