摘要
Cohesion is an important soil strength parameter for the overall structure and quality of building foundations as well as slope stability. For a civil engineering project, cohesion (c) can be determined directly from mainly unconfined compression tests, direct shear tests, and triaxial tests of soil. However, it’s quite challenging to collect soil samples as there are time and cost constraints, as well as a diversity of soil deposits. Hence, this research aims to demonstrate a simplified method in order to determine the strength parameter of cohesive soil. Here, we propose an alternative solution adopting statistical correlations and machine learning techniques to establish correlations between the liquid limit, plastic limit, moisture content and %fine of soil with the strength parameter. In laboratory testing, these parameters can be obtained easily and these tests are relatively simple, quick to perform and also comparatively inexpensive. Hence, several test results were used from 100 boreholes which were soft soil or silty clay-type soil. Using the collected in-situ and lab test results of soil samples, a Multiple Linear Regression (MLR), Random Forest Regression (RFR) and Machine Learning (ML) model will be developed to establish a relationship between cohesion and the available test results. In order to assess the performances of both models, several performance indicators like: correlation coefficient (R<sup>2</sup>), mean squared error (MSE), root mean square error (RMSE), and mean average error (MAE) will be used. These correlation coefficients will be used to demonstrate the prediction capacity and accuracy of both models. It should be noted that this approach will substitute the conventional testing required for strength parameters, which is both expensive and time-consuming.
Cohesion is an important soil strength parameter for the overall structure and quality of building foundations as well as slope stability. For a civil engineering project, cohesion (c) can be determined directly from mainly unconfined compression tests, direct shear tests, and triaxial tests of soil. However, it’s quite challenging to collect soil samples as there are time and cost constraints, as well as a diversity of soil deposits. Hence, this research aims to demonstrate a simplified method in order to determine the strength parameter of cohesive soil. Here, we propose an alternative solution adopting statistical correlations and machine learning techniques to establish correlations between the liquid limit, plastic limit, moisture content and %fine of soil with the strength parameter. In laboratory testing, these parameters can be obtained easily and these tests are relatively simple, quick to perform and also comparatively inexpensive. Hence, several test results were used from 100 boreholes which were soft soil or silty clay-type soil. Using the collected in-situ and lab test results of soil samples, a Multiple Linear Regression (MLR), Random Forest Regression (RFR) and Machine Learning (ML) model will be developed to establish a relationship between cohesion and the available test results. In order to assess the performances of both models, several performance indicators like: correlation coefficient (R<sup>2</sup>), mean squared error (MSE), root mean square error (RMSE), and mean average error (MAE) will be used. These correlation coefficients will be used to demonstrate the prediction capacity and accuracy of both models. It should be noted that this approach will substitute the conventional testing required for strength parameters, which is both expensive and time-consuming.
