<p>The substantial time and financial costs associated with traditional soil mechanics testing pose significant challenges in characterizing the mechanical properties of completely weathered granite (CWG), a distinctive soil type. This study proposes a machine learning (ML) framework to enable efficient and accurate prediction of the mechanical behavior of CWG soils, based on limited experimental data. First, a series of consolidation drainage tests were conducted to investigate the mechanical properties of CWG in a fault zone. The experimental results reveal that CWG exhibits pronounced moisture sensitivity, with shear strength and failure modes depending more strongly on moisture content than on dry density or confining pressure. Notably, as moisture content increases from near-dry to saturated conditions, shear strength follows a nonlinear trend—initially decreasing rapidly before gradually recovering—while four distinct failure modes emerge. Based on experimental results, a total of six machine learning (ML) models were evaluated by cross-validation and comparisons with experiments, among which the XGB model possesses superior predictive accuracy and smallest dispersion. Finally, a user-friendly interactive web application (APP) was established to support the practical implementation of the prediction model for CWG soils. The core contribution of this paper lies in understanding the unique mechanical properties of CWG soil, and providing an intelligent prediction methodology.</p>