This study evaluates the applicability of using a robust, novel, data-driven method in proposing surrogate models to predict the maximum dry unit weight, optimum moisture content, and California bearing ratio of coarse-grained soils using only the results of the grain size distribution analysis. The data-driven analysis has been conducted using evolutionary polynomial regression analysis (MOGA-EPR), employing a comprehensive database. The database included the particle diameter corresponding to a percentage of the passing of 10%, 30%, 50%, and 60%, coefficient of uniformity, coefficient of curvature, dry unit weight, optimum moisture content, and California bearing ratio. The statistical assessment results illustrated that the MOGA-EPR provides robust models to predict the maximum dry unit weight, optimum moisture content, and California bearing ratio. The new models’ performance has also been compared with the empirical models proposed by different researchers. It was found from the comparisons that the new models provide enhanced accuracy in predictions as these models scored lower mean absolute error and root mean square error, mean values closer to one, and higher and coefficient of correlation. Therefore, the new models can be used to ensure more optimised and robust design calculations.
Read full abstract