Predictive Genetic Programming Approaches for Swell-Shrink Soil Compaction

Genetic programming (GP) is a machine learning tool to predict the maximum dry density (rho(dmax)) as well as optimum water content (OMC) of expansive soils ('rho dmaxOMC-ES') in Sharjah city by using a comprehensive experimental database comprising 311 points with the help of gene expression programming (GEP) and multi-gene expression programming (MEP) approaches. Mathematical expressions were simplified to compute the rho dmaxOMC-ES for both genetic models. A variety of error indices, i.e., mean absolute error, root mean square error, Nash-Sutcliffe efficiency, and coefficient of correlation (R), experimental to predicted ratios, alongside the sensitivity and monotonicity analyses, were utilized to assess the proposed models' efficacy. The findings demonstrate that rho dmaxOMC-ES can be robustly characterized using GEP and MEP methods, yielding higher prediction performance; however, the GEP model produced comparatively superior accuracy (R-TrD(2) = 0.81 and 0.67, R-TsD(2) = 0.84 and 0.68, for rho(dmax) and OMC, respectively). Furthermore, when the suggested genetic models were compared with past models, these performed more efficaciously and robustly. The rho(dmax) models yielded substantially improved results compared to the OMC predictive GEP and MEP models. Both GP-based models are efficacious in determining the rho dmaxOMC-ES in geo-environmental engineering, thereby reducing time and labour-intensive testing and promoting sustainability. The formulation of these GP-based forecasting models for evaluating the compaction parameters of swell-shrink soils is a desideratum to enhance the sustainability and resilience of transportation infrastructure against the challenges posed by these calamitous soils.