Regression Analysis for Predicting Soil Strength in Bangladesh

Abstract

This study focuses on establishing a robust relationship between Standard Penetration Test-N values (SPT-N), geotechnical parameters and unconfined compressive strength (qu) using regression analysis. The proposed relationship offers a reliable method for estimating qu based on SPT-N values. A comprehensive dataset comprising approximately 200 soil samples collected from various boreholes across Dhaka city was utilized. Multiple Linear Regression (MLR), Rando-forest Regression (RFR) and AdaBoost Regression techniques were employed to develop a unified correlation model. Evaluation metrics including R-squared (R2), Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), along with Trend-behavior Analysis were employed to assess and compare the performances of the models. Additionally, sensitivity analysis was carried out on the selected model in order to assess the importance of each parameter used to predict qu. Finally, the selected model was compared against the existing empirical models that were published in previous studies. In terms of evaluation metrics and Trend-behavior Analysis, the results showed that the RFR model performed better than the others. Additionally, the selected model outperformed the others, demonstrating the highest R2 score, the smallest RMSE and MAE values and lower residuals compared to the previous models. Hence, the proposed model provides accurate predictions of qu for clayey soil in Bangladesh. Its implementation could ensure more efficient geotechnical designs, specifically adjusted to the geological conditions of the Dhaka region. While previous studies have established regional equations for various parts of the world, our model uniquely has incorporated the Plasticity Index (PI) as a predictor for qu and is specifically calibrated for the geological characteristics of Dhaka city. The findings of this study highlight the effectiveness and applicability of regression analysis in predicting qu for Dhaka's soil properties, thus introducing a valuable tool for enhancing the accuracy and effectiveness of geotechnical assessments and design in the region. KEYWORDS: Unconfined compressive strength, Standard penetration test-N values, Plasticity index, Multiple linear regression, Random-forest regression, AdaBoost regression, Evaluation metrics, Trend-behavior analysis, Sensitivity analysis