Estimating the strength of soil stabilized with cement and lime at optimal compaction using ensemble-based multiple...

Area of Science:

• Geotechnical Engineering
• Environmental Geotechnics
• Materials Science

Background:

• Cohesive soils require stabilization for pavement subgrades and landfill liners when unconfined compressive strength (UCS) is below 200 kN/m².
• Improving mechanical properties is crucial for structural integrity and environmental protection in geotechnical applications.
• Machine learning offers advanced analytical tools for predicting soil behavior and optimizing material properties.

Purpose of the Study:

• To comparatively assess machine learning models for predicting the unconfined compressive strength (UCS) of cohesive soil stabilized with cement and lime.
• To identify the most effective ensemble-based machine learning techniques for soil stabilization analysis.
• To determine the key factors influencing the UCS of reconstituted cohesive soils.

Main Methods:

• Utilized ensemble-based machine learning classification (Gradient Boosting, CN2, Naïve Bayes, SVM, SGD, K-NN, Decision Tree, Random Forest) and symbolic regression (ANN, RSM).
• Trained and tested models on 190 experimental data points, considering inputs: cement, lime, liquid limit, plasticity index, optimum moisture content, and maximum dry density.
• Performed correlation matrix and sensitivity analysis to identify influential parameters on UCS.

Main Results:

• Gradient Boosting (GB) and K-Nearest Neighbors (K-NN) models achieved the highest accuracy (95%).
• CN2, Support Vector Machine (SVM), and Decision Tree (Tree) models showed approximately 90% accuracy.
• Maximum dry density (MDD), consistency limits (LL, PI), and cement content significantly influenced UCS, while optimum moisture content (OMC) had a negligible impact.

Conclusions:

• Ensemble machine learning models, particularly GB and K-NN, are highly effective for predicting the UCS of stabilized cohesive soils.
• Optimal soil stabilization can be achieved by focusing on MDD, consistency limits, and cement content.
• The findings provide a valuable framework for field applications in designing stable geotechnical structures with reconstituted soils.