Using machine learning technique for designing reinforced lightweight soil

Abstract

Taking advantage of dredged sediments as lightweight materials is a useful solution to protect the environment and save natural materials in the field of construction. In which unconfined compression strength is an important criterion to determine the application in the construction project. It is difficult to find the optimal mixing ratio based on design standards or construction conditions because the unconfined compression strength (UCS) is affected by the mixing ratio of the materials and additives. In this study, the Machine Learning (ML) models consisting of Extreme Gradient Boosting (XGB) model and Linear regression models are investigated to design components for reinforced lightweight soil based on the influence of unconfined compression strength of the test sample which is water content, cement content, air foam content, waste fishing net. To evaluate the effectiveness of the proposed ML models, several evaluation criteria including Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and coefficient of determination (R2) are proposed. The results show that the predictions of the XGB model have high accuracy with R2 = 0.9695, RMSE = 5.5849 kPa and MAE = 4.1583 kPa for the testing dataset. Sensitivity analysis evaluates the influence of input variables on UCS and the interaction between input variables to help design RLS components optimally.