Determining Relations Between SPT-N & Shear Strength Parameters of Dhaka and Sylhet soils using Machine Learning Approach

Abstract

The study is about establishing relationship between SPT N values, geotechnical parameters of soil & Angle of Internal Friction (φ), unconfined compressive strength (qu) for the region of Dhaka City & Sylhet region of Bangladesh using Machine learning technique. The relationship represents a formula for estimating Angle of Internal Friction φ & unconfined compressive strength qu from SPT N value. The relationship was formed previously by other researchers for regions of USA, Japan, Malaysia, UK and India. For Bangladesh a study was performed for areas of Joydevpur, Mymensingh, Jamalpur, Dhaka metro, Tangail & Madhupur. Though many countries have their own regional equation, for angle of internal friction (φ) finer content was not applied and for unconfined compressive strength (qu) plasticity index was not applied in empirical equations, the study areas were also different. For this study About 300 samples have been collected from boreholes from Mirpur, Uttara, 300ft, and Purbachal area of Dhaka City & more than 400 samples were collected from Sylhet, Narsingdi, Habiganj, Brahmanbaria locations of Sylhet region. To develop the relation model & estimation linear regression, Multi Linear Regression (MLR), Structure Vector Regression (SVR) algorithms was used. According to R2 , RMSE & MSE value the best correlation is chosen among several combinations of N, N60, N1,60, depth & grain size data for Angle of Internal Friction (φ) and N, N60, depth & plasticity data for unconfined compressive strength (qu). The best relation has later been compared with SVM model values where in some cases the MLR model comes out as better in terms of R2 , RMSE & MSE value and in some cases the SVM model comes out as better one in terms of R2 , RMSE & MSE value. Then the predicted values from selected MLR & SVM model have been compared with previously established empirical equations where the model shows better values of R2 , RMSE & MSE than previous established models. The better values from evaluation matrices indicates the better predicting ability of Angle of Internal Friction φ & unconfined compressive strength qu for the soils of Dhaka City & Sylhet region of Bangladesh.