Machine Learning–Informed Geomaterial Design for Embankment Construction

This paper investigated standalone and hybrid machine learning methods for predicting the angle of internal friction in frictional and cohesive frictional soils, essential properties in embankment construction. A comprehensive laboratory analysis of 149 samples collected from a highway project was conducted to determine grain size, plasticity, compaction characteristics, and friction angle. Factors such as gravel and sand content, fines content, liquid limit, plasticity index, maximum dry density, and optimal moisture content were used as input variables for modelling. The data set was enhanced using wavelet decomposition by applying discrete wavelet transform (WT), which was then integrated into a white shark optimizer (WSO)–assisted regression model. The WT-WSO model was found to outperform traditional regression approaches. Comparative findings indicated that the WT-WSO model achieved superior performance in predicting the friction angle compared to standalone WSO methods. Finally, the machine learning models optimized backfill selection, improving the efficiency and reliability of earth-retaining system designs.