TITLE:
Prediction of Strength Properties of Soft Soil Considering Simple Soil Parameters
AUTHORS:
Md. Janibul Hoque, Md. Bayezid, Ahnaf Rafi Sharan, Mozaher Ul Kabir, Tahsin Tareque
KEYWORDS:
Cohesion, MLR, RFR, ML, MSE, RMSE, MAE
JOURNAL NAME:
Open Journal of Civil Engineering,
Vol.13 No.3,
September
8,
2023
ABSTRACT: 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 (R2), 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.