Application of TreeNet® Regression for Generating a Standard Line in Soil Compaction Testing

Abstract

This research aims to develop and evaluate a machine learning framework for predicting Dry Density (DD) based on Moisture Content (MC), a fundamental relationship in geotechnical engineering. Utilizing a dataset of 600 samples, a tree-based ensemble model was developed and optimized through 5-fold cross-validation. The model's predictive performance was quantified using the coefficient of determination (R²) and Mean Absolute Percentage Error (MAPE). The optimized model, comprising 296 trees, achieved an R² of 48.73% and a MAPE of 12.14% on the test dataset. Results indicate that the model generalized well to unseen data without exhibiting signs of overfitting. Notably, the ensemble approach effectively captured the complex, non-linear, and non-monotonic relationship between MC and DD, demonstrating that the sensitivity of dry density to moisture fluctuations varies significantly across different ranges. This study concludes that machine learning techniques offer a robust alternative for modeling intricate geotechnical behaviors where traditional linear methods may fail, reaffirming moisture content as a primary predictor of soil compaction outcomes.