ANN-based swarm intelligence for predicting expansive soil swell pressure and compression strength

Area of Science:

• Geotechnical Engineering
• Computational Intelligence
• Materials Science

Background:

• Expansive soils pose significant challenges in civil engineering projects due to their volume change behavior.
• Accurate prediction of swell pressure and unconfined compression strength is crucial for foundation design.
• Existing prediction models often struggle with the complex interplay of factors influencing expansive soil behavior.

Purpose of the Study:

• To develop and evaluate novel artificial neural network (ANN) models integrated with metaheuristic optimization algorithms for predicting swell pressure and unconfined compression strength of expansive soils (P_sUCS-ES).
• To compare the performance of ANN models optimized with Particle Swarm Optimization (PSO), Grey Wolf Optimization (GWO), Slime Mould Algorithm (SMA), and Marine Predators' Algorithm (MPA).
• To identify the most effective model for accurate and reliable prediction of P_sUCS-ES.

Main Methods:

• Four ANN-based models (ANN-PSO, ANN-GWO, ANN-SMA, ANN-MPA) were developed using nine influential parameters derived from 145 published papers.
• Model performance was assessed using metrics like MAE, R², RMSE, RSR, VAF, WI, and WMAPE.
• Sensitivity and monotonicity analyses were conducted to validate model robustness.

Main Results:

• All developed ANN models achieved a regression coefficient (R) > 0.8 for the overall dataset.
• The ANN-MPA model demonstrated superior performance, yielding the highest R values across training, testing, and validation datasets and the lowest Mean Absolute Error (MAE).
• While some models showed overfitting in unconfined compression strength (UCS) predictions, ANN-MPA exhibited robust performance without significant overfitting, with most predictions falling within ±20% error.

Conclusions:

• Swarm-based ANN models, particularly ANN-MPA, offer a robust solution for predicting P_sUCS-ES, effectively handling hyperparameter tuning complexities.
• The ANN-MPA model shows high potential for practical application in geoenvironmental engineering due to its accuracy and reliability.
• The findings support the integration of advanced computational intelligence techniques for improved geotechnical engineering predictions.