Tuesday 04 March 2025
Scientists have made a significant breakthrough in predicting the strength of subgrade soil, a crucial component in building construction and infrastructure development. Researchers from several institutions collaborated on a study that used advanced machine learning techniques to forecast the California Bearing Ratio (CBR), Unconfined Compressive Strength (UCS), and Resistance Value (R) of modified subgrade soil.
The team employed three machine learning algorithms – CatBoost, XGBoost, and Support Vector Regression (SVR) – to analyze a dataset of 121 experimental samples. Each sample consisted of varying proportions of hydrated lime activated rice husk ash (HARSH), along with other factors such as plastic limit, liquid limit, and clay activity.
The results showed that the XGBoost algorithm outperformed the others in predicting CBR, UCS, and R values. Specifically, the model achieved a coefficient of determination (R2) of 0.9999 for CBR, 0.9995 for UCS, and 0.9997 for R.
The researchers also conducted sensitivity analysis to understand how each input feature affected the model’s predictions. They found that increasing the proportion of HARSH in the soil led to higher predicted values for all three mechanical properties. This is an important finding, as it suggests that using more HARSH can improve the strength and stability of subgrade soils.
The study also compared the performance of their models with previous research. The XGBoost-based model outperformed existing methods in predicting CBR and R values, while showing comparable results for UCS prediction.
This breakthrough has significant implications for the construction industry. Accurate predictions of subgrade soil strength can help engineers design safer and more durable infrastructure projects. Moreover, this study demonstrates the potential of machine learning algorithms in solving real-world problems that were previously challenging to tackle.
The researchers hope that their findings will be used to develop new methods for predicting the mechanical properties of subgrade soils. They also plan to extend their research to other types of soil and to explore the use of additional machine learning techniques.
In practical terms, this study has the potential to save time and money by reducing the need for costly and time-consuming laboratory testing. It can also help ensure that infrastructure projects are designed with a higher degree of confidence and accuracy.
Overall, this research marks an important step forward in the application of machine learning to civil engineering problems.
Cite this article: “Machine Learning Breakthrough Predicts Strength of Subgrade Soil”, The Science Archive, 2025.
Machine Learning, Civil Engineering, Subgrade Soil, California Bearing Ratio, Unconfined Compressive Strength, Resistance Value, Harsh, Xgboost, Svr, Catboost
