Predicting soil compaction parameters in expansive soils using advanced machine learning models: a comparative study.

Expansive soils present substantial challenges for construction projects, particularly in arid regions where their swelling and shrinking behavior can result in structural damage. This research conducted an in-depth analysis of 195 soil samples, investigating six crucial factors, including clay content, liquid and plastic limit, specific gravity, plasticity index, and swell percentage, influencing soil compaction. The study used five computational models to predict optimal water content and maximum dry density. Among the models, XG-Boost exhibited exceptional performance with high predictive accuracy, scoring 0.941 and 0.912 for water content and density, respectively. Random Forest method also performed well, whereas Long Short-Term Memory Network (LSTMN) and k-Nearest Neighbors methods achieved moderate success. Detailed performance metrics, including correlation coefficients (r), mean absolute error (MAE), root mean squared error (RMSE), relative root mean squared error (RRMSE), and performance index (PI), were analyzed for each model across the training, testing, and validation datasets. XG-Boost consistently outperformed other models, demonstrating its robust and reliable predictive power. The analysis of absolute error highlighted the accuracy and consistency of XG-Boost and Random Forest in predicting the maximum dry density.