Modeling resilient modulus of subgrade soils using LSSVM optimized with swarm intelligence algorithms.

Area of Science:

• Geotechnical Engineering
• Pavement Engineering
• Soft Computing
• Machine Learning

Background:

• Resilient modulus (Mr) is critical for pavement design, but laboratory determination is often impractical due to soil variability and testing complexities.
• Accurate prediction of subgrade soil Mr is essential for reliable pavement structural analysis and performance.
• Traditional methods for Mr determination face challenges with spatial variability and testing protocols, necessitating advanced predictive techniques.

Purpose of the Study:

• To develop and evaluate a novel soft computing approach for predicting the resilient modulus (Mr) of subgrade soils.
• To compare the efficacy of hybrid Least Square Support Vector Machine (LSSVM) models optimized by various swarm intelligence algorithms.
• To identify the most influential soil parameters for accurate Mr prediction in pavement engineering.

Main Methods:

• Hybridization of Least Square Support Vector Machine (LSSVM) with six swarm intelligence algorithms: Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), Symbiotic Organisms Search (SOS), Salp Swarm Algorithm (SSA), Slime Mould Algorithm (SMA), and Harris Hawks Optimization (HHO).
• Utilized a literature dataset comprising 891 soil samples with input variables including stress, unconfined compressive strength, moisture content, and particle size distribution.
• Assessed model accuracy using statistical metrics such as Root Mean Square Error (RMSE) and Coefficient of Determination (R²).

Main Results:

• Percent passing No. 200 sieve, optimum moisture content, and unconfined compressive strength were identified as the most significant predictors of Mr.
• The hybrid models LSSVM-GWO, LSSVM-SOS, and LSSVM-SSA demonstrated superior performance in predicting Mr compared to other tested algorithms.
• LSSVM-SOS achieved the highest accuracy with RMSE of 6.72 MPa and R² of 0.942, closely followed by LSSVM-SSA (RMSE 6.78 MPa, R² 0.940) and LSSVM-GWO (RMSE 6.79 MPa, R² 0.940).

Conclusions:

• The developed hybrid LSSVM models, particularly LSSVM-SOS, offer a highly accurate and efficient alternative for predicting subgrade soil resilient modulus.
• Swarm intelligence optimization significantly enhances the predictive capability of LSSVM for geotechnical engineering applications in pavement design.
• The findings suggest that LSSVM-SOS is a reliable tool for high-accuracy estimation of Mr, supporting improved pavement structural design.