Tree-Based Techniques for Predicting the Compression Index of Clayey Soils

Abstract

Compression index is an effective assessment of primary consolidation settlement of clayey soils, but the process of obtaining compression index is time-consuming and laborious. Thus, in the present study, we developed two classical tree-based techniques: random forest (RF) and extreme gradient boosting (XGBoost), to predict the compression index of clayey soils. To establish these two models, we collected an available dataset—including 391 consolidation tests for soils—from previously published research. The dataset consists of six physical parameters, including the initial void ratio, natural water content, liquid limit, plastic index, specific gravity, and soil compression index. The first five parameters are the models’ inputs while the compression index is the models’ output. We trained both two tree-based models using 90% of the entire dataset and used the remaining 10% to assess the well-trained models, which is consistent with the published research. Several statistical metrics, such as coefficient of determination (R2), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), are the criteria for assessing the models’ performance. The results show that the RF model has better accuracy in predicting compression index compared with the XGBoost model because it outperforms the XGBoost model both on the training and testing datasets. The performance of the RF model is R2 of 0.928 and 0.818, RMSE of 0.016 and 0.025, MAPE of 7.046% and 10.082%, and MAE of 0.012 and 0.020 on the training and testing datasets, respectively. The sensitivity analysis reveals that the initial void ratio has a significant impact on the compression index of clayey soils.

Keywords

• compression index
• consolidation
• machine learning
• random forest
• extreme gradient boosting