Interpreting Non‐Linear Compaction Mechanics of Expansive Soils Using Reliability Calibrated Framework

ABSTRACT

Maximum dry density (MDD) and optimum moisture content (OMC) are fundamental parameters for compaction design and construction. Existing data driven studies treat these parameters as deterministic prediction targets and select models mainly from point estimate accuracy which limits their value for engineering decisions where uncertainty and reliability bounds are important. This study develops a reliable framework for estimating MDD and OMC in expansive soils using index properties. Gaussian Process Regression provided the highest MDD accuracy with R2 = 0.838 and RMSE = 60.84 kg/m ³ , while CatBoost produced the best OMC accuracy with R2 = 0.603 and RMSE = 1.844%. The results show a clear difference between the predictability of MDD and OMC indicating that dry density is better represented by routine index properties, whereas moisture demand remains more sensitive to mineralogical controls that are only captured by LL and PI. Jackknife+ and cross‐conformal prediction produced 90% prediction intervals, with observed coverages of approximately 89% for MDD and 87% for OMC, allowing each estimate to be accompanied by practical uncertainty bounds. The proposed framework therefore extends compaction modeling beyond deterministic accuracy comparison by combining multi model reliability screening, MDD and OMC estimation, and uncertainty for initial design and construction quality control.