Effective Machine Learning Models for Predicting SPT N of Reclaimed Jolshiri Area, Dhaka

Accurate prediction of soil properties, particularly the Standard Penetration Test (SPT) N value, is crucial for urban multistory constructions. However, predicting SPT N values can be challenging in areas with diverse soil types and unique geological formations, such as the Jolshiri Abashon reclaimed land. To address this challenge, this study integrated existing field test data and utilized machine learning algorithms to create a prediction model for SPT N values and prepare a heatmap to visually represents the distribution and changes of the SPT N value with depth and other parameters. The study identified several parameters, including effective stress, depth, latitude, longitude, and soil type, that significantly influence the accurate prediction of SPT N values. Among the machine learning models employed, the Random Forest model exhibited the best performance, achieving a high R2 value of 0.983, a low Mean Absolute Error (MAE) of 1.867, and a Root Mean Squared Error (RMSE) of 2.27. The decision tree model also performed well, with an R2 value of 0.979, an MAE of 1.2528, and an RMSE of 2.53. However, the AdaBoost model demonstrated poorer performance in this context. Overall, this study highlights the importance of utilizing geospatial data and machine learning algorithms to accurately predict SPT N values in challenging areas like Jolshiri Abashon. The results demonstrate the potential for these techniques to improve the precision and effectiveness of estimation in soil engineering projects, ultimately leading to safer and more reliable constructions.