Geotechnical parameter estimation based on machine learning techniques using three dimensional soil maps

In this thesis study, the aim is to model local soil characteristics digitally for the Fatih district of Istanbul and to estimate geotechnical parameters based on these models using machine learning techniques. The study consists of three main stages. In the first stage, data obtained from soil and geotechnical investigation reports prepared at different times were transferred to a digital media to create a comprehensive and organized database. This database includes both field investigation results and laboratory test results. In the second stage, the study area was mapped in detail using open-source geographic information system (GIS) software such as Google Earth and QGIS; subsequently, three-dimensional soil models were created using the CAD and GIS-based NETPROMine software. By this way, the spatial distribution and engineering properties of soil layers were visualized in detail. In the third and final stage, the data obtained from these models were analyzed using artificial neural network (ANN) algorithms in the Matlab software, and the estimation of consistency and strength parameters was performed. Regression analyses, multivariate data evaluations, and various error metrics were used in the estimation processes. The prediction results yielded low error values, and the reliability of the models was tested through comparisons with existing literature. The results contribute to the rapid, reliable, and sustainable estimation of geotechnical parameters, especially in urban areas with complex structures where field investigations are limited. The study presents an innovative and digital-based approach for engineering applications. The generated three-dimensional soil maps help visualize subsurface conditions, facilitating the decision-making processes of field engineers. The modeling process presents an alternative to traditional drilling methods, especially in archaeologically and architecturally sensitive areas. In this respect, the study is not limited to the Fatih district and serves as an example for other urban areas with similar characteristics.