•  
  •  
 

Abstract

This study suggests that RF and ANN are proven to be robust algorithms in predicting in-situ soil density, which is considered a significant geotechnical parameter. The research is based on 86 soil samples and focuses on five main input parameters: Gravel Percentage (G%), Plastic Limit (PL%), Sand Percentage (S%), Fines Percentage (F%), and Liquid Limit (LL%). The models developed here utilize five commonly recorded index properties (G%, S%, F%, LL, and PL) for all field samples taken from the Basra-Faw Road project. The influence of moisture content and compressive energy was ignored, as all field samples acquired the same moisture content and compressive energy during compression. The fit of the two models was thoroughly tested with statistical indices, including the coefficient of determination (R2) and Root Mean Square Error (RMSE). The analysis shows that the ANN model has better predictive performance compared to the RF model, with the R2 and the RMSE equal to 0.98786 and 0.0027 for the ANN model and 0.96249 and 0.0192 for the RF model. This result emphasizes the ANN's great capability in capturing the complicated non-linear relationship between input variables and soil density. Moreover, the study reveals gravel and fines percentages as the most significant parameters that control the prediction of soil density. Results indicated that machine learning methods, namely ANN, can be an easy, quick, and nondestructive alternative to traditional field-testing methods to predict soil compaction. The research findings add to the base of the art in geotechnical engineering by highlighting the benefits of advanced predictive tools in improving soil density revocation accuracy and efficiency. Incorporating other factors, such as moisture content and compressive energy during compaction, into future datasets may enhance the model's generalizability and accuracy. The findings of this study can have significant implications for bidding purposes and safety in infrastructure-related design; the accuracy of the soil density predictions is critical to such applications as foundation design, slope protection, and pavement construction.

Keywords

Random forest (RF), Artificial neural networks (ANN), In-situ soil density, Machine learning, Geotechnical engineering, Predictive modeling

Download

Share

COinS