Predicting and evaluating settlement of shallow foundation using machine learning approach.

Area of Science:

• Geotechnical Engineering
• Computational Mechanics
• Civil Engineering

Background:

• Accurate prediction of shallow foundation settlement is essential for structural integrity and safety.
• Machine learning offers promising tools for geotechnical analysis, but model optimization requires careful evaluation.
• Understanding the influence of various parameters on foundation settlement is critical for design.

Purpose of the Study:

• To evaluate the effectiveness of Particle Swarm Optimization (PSO) in enhancing machine learning models for shallow foundation settlement prediction.
• To compare the performance of Gradient Boosting (GB), Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) models, both with and without PSO tuning.
• To identify the key variables influencing shallow foundation settlement through sensitivity analysis.

Main Methods:

• Development and evaluation of four hybrid machine learning models: GB-PSO, RF-PSO, SVM-PSO, and KNN-PSO.
• Utilizing an experimental dataset of 189 samples for model training and validation.
• Rigorous performance assessment using K-Fold Cross-Validation, R², RMSE, MAE, and MAPE metrics.
• Sensitivity analysis employing Shapley Additive Explanation (SHAP) to determine variable importance.

Main Results:

• Particle Swarm Optimization (PSO) did not consistently improve the prediction accuracy of the machine learning models.
• Original Gradient Boosting (GB) and Random Forest (RF) models demonstrated superior performance compared to their PSO-optimized counterparts.
• Average Standard Penetration Test (SPT) blow count and footing width (B) were identified as the most significant variables affecting settlement predictions.
• Footing embedment ratio (Df/B) and net applied pressure (q) also showed considerable influence on settlement predictions.

Conclusions:

• Standard machine learning models, particularly GB and RF, are highly effective for predicting shallow foundation settlement without PSO enhancement.
• The Standard Penetration Test (SPT) blow count and footing width (B) are critical parameters that engineers must consider in settlement analysis.
• A user-friendly Excel tool based on the GB model is provided for practical application in civil engineering, aiding in reliable settlement predictions.