Establishing the co-relations between different geotechnical parameters of Bangladesh coastal soil using machine learning techniques

Abstract

The digital revolution is currently leaving no sector untouched. The combination of data and digital technologies opens up a multitude of opportunities in the geotechnical sector and Machine Learning is undeniably one of the most innovative applications in predicting soil parameters. Even so, owing to the uncertainty regarding the accuracy of the prediction models, traditional methods of determining soil parameters are still being used, which are both costly and time consuming. The purpose of this study is to correlate the different soil parameters of Bangladesh coastal soil, such as SPT N value, Shear wave velocity, Fine Content, Cohesion, and stiffness, and then use the correlation to predict the Angle of Friction. To predict the angle of friction of the coastal soil, six machine learning techniques were used: Simple linear regression model, Multi polynomial Regression, Support Vector Regression, Random Forest, Multivariate Adaptive Regression Splines, M5 Model Tree, and Artificial Neural Network. About 58 data sets were collected and used for this research project. Among 58 data sets, 48 were used to correlate the soil parameters and 10 data sets were used for testing and validation. Furthermore, all the machine learning methods were compared in terms of prediction accuracy. Finally, a validation of the predicted result has been conducted using PLAXIS 2D Software. In general, the Random Forest and M5 model tree regression models generated the best results, as the R2 value (97.95% & 95.23% respectively) is the highest among all of the models and the error-values are also lower, reflecting better accuracy. Moreover, it is more evident from the study that conventional machine learning technique shows better performance than ANN where there is data scarcity.