An evolutionary optimized automated machine learning approach to soil unconfined compressive strength prediction for sustainable transportation infrastructure

Abstract

Soil chemical stabilization recommendations use treated soils’ unconfined compressive strength (UCS) as the main acceptance criterion in laboratory tests. However, optimizing UCS supplemental content requires a human- and financial-intensive trial-and-error process. Data intelligence models enhance automatized scientific sampling procedures, limit laboratory testing, and provide useful information regarding stabilization adequacy without producing preliminary samples. This research proposes an evolutionary algorithm-assisted automated gradient boosting model to predict the UCS values from datasets from diverse sources. A grey wolf optimization algorithm is integrated into the gradient boosting training procedure, determining its best internal parameters and helping to select the most relevant input variables. Comparative evaluations on six recently published datasets demonstrate the efficiency of the proposed model compared to existing approaches. The optimized models produced better results than the benchmark models reported in the literature, with average coefficients of determination ranging from 0.723 to 0.928. The hybrid models with evolutionary feature selection achieved comparable performance while reducing the number of input variables between 16% and 54%.