A super-learner machine learning model for a global prediction of compression index in clays

Abstract

Settlement of structures is determined by the stiffness of the soil where they are built. Compression index (cc) quantifies the compressibility of the soil and is a key parameter in the design of geotechnical structures. To predict the value of cc in clay soils, a global database of more than 1000 data points was collected and analysed. Liquid limit, plasticity index, natural water content, and initial void ratio were considered as predictor variables. A super-learner machine learning model was developed to predict cc from these variables. The model demonstrated a reasonable predictive performance and was subsequently integrated into an online tool. Additionally, four symbolic regression expressions were obtained to relate cc with some of the input variables when not all data are available, providing simple and practical alternatives for cc, estimation. This study provided two major contributions: (1) the non-local nature of the data expands the scope and generalizability of the findings, and (2) the availability of the proposed algorithm through an online application ensures its accessibility for geotechnical engineers, enhancing the work’s practical applicability and intrinsic value.