Unconfined compressive strength prediction of stabilized expansive clay soil using machine learning techniques

• Main Authors: Ahmad, Mahmood, Al-Mansob, Ramez Al-Ezzi Abduljalil, Ramli, Ahmad Bukhari, Ahmad, Feezan, Jehan Khan, Beenish
• Format: Article
• Language: English; English
• Published: Springer Nature 2024
• Subjects: TA Engineering (General). Civil engineering (General); TA401 Materials of engineering and construction
• Online Access: http://irep.iium.edu.my/106975/1/106975_Unconfined%20compressive%20strength.pdf; http://irep.iium.edu.my/106975/7/106975_Unconfined%20compressive%20strength_SCOPUS.pdf; http://irep.iium.edu.my/106975/; https://link.springer.com/article/10.1007/s41939-023-00203-7
• Institution: Universiti Islam Antarabangsa Malaysia

This paper evaluates the potential of machine learning techniques, namely, Gaussian Process Regression (GPR) and Support Vector Machine (SVM), for the prediction of unconfined compressive strength (UCS) of expansive clay soil treated with hydrated-lime-activated rice husk ash. A laboratory dataset containing 121 records has been used with input parameters, including hydrated-lime-activated rice husk ash, liquid limit, plastic limit, plastic index, optimum moisture content, clay activity, and maximum dry density. The performances of the GPR and SVM models are assessed using statistical metrics, including the coefficient of determination (R2), mean absolute error (MAE), root-mean-square error (RMSE), relative rootmean-square error (RRMSE), and performance indicator (ρ). The analysis of the R2 together with MAE, RMSE, RRMSE, and ρ values for the UCS demonstrates that the SVM and GPR models achieved better prediction results, i.e., R2 0.9998, MAE 0.0514, RMSE 0.1408, and ρ 0.0004 and R2 0.9998,MAE 0.3430, RMSE 0.4455, and ρ 0.0011, respectively, as compared to the artificial neural network model recently developed in the literature with (R2 0.9900, MAE 0.3500, RMSE 4.9300, RRMSE 0.2000, and ρ 0.1000) in test phase, which indicates that both models are efficient and reliable for practical applications. Furthermore, the sensitivity analysis result shows that maximum dry density was the key parameter affecting the UCS.