Jordan Journal of Civil Engineering

Regression Analysis for Predicting Soil Strength in Bangladesh


Shadman Rahman Sabab; Hossain Md. Shahin; Md. Muftashin Muhim Bondhon; Md. Ehsan Kabir;


The study is about establishing a relationship between SPT-N values, geotechnical parameters of soil, and unconfined compressive strength (qu) for Dhaka City, Bangladesh region using Machine learning algorithms. The relationship represents a method for estimating unconfined compressive strength (qu) from the SPT-N value. For this study, about 200 samples have been collected from boreholes in different parts of Dhaka. Multiple Linear Regression (MLR), Random Forest Regression, and AdaBoost regression were carried out to develop a unified correlation. Evaluation metrics: R2, RMSE & MAE values were used to assess and compare the models. Then the selected model is compared with empirical models that were published in previous studies. It was found that Random Forest Regression (RFR) performed better by producing the largest R2 score, smallest RMSE, and MAE value compared to the others and also this chosen model has significant results in evaluation metrics and comparatively low residuals compared to previous models. Thus, this model can predict the unconfined compressive strength (qu) of the clayey soil of Bangladesh with substantial accuracy. This model can ensure more efficient geotechnical designs. This type of relationship was formed previously by other researchers for a specific region of various parts of the world. Though many countries have their regional equation, the plasticity index was not included in their equations of qu, and the study areas were also different however they are used more frequently in the Dhaka region. Thus, this model could be more convincing in the aspect of Dhaka’s geological condition.


Clayey soil, SPT-N60, Unconfined Compressive Strength (qu), Plasticity Index, Machine-Learning Algorithms