This study focuses on the investigation of detecting soil profiles encountered during exploratory drilling using continuous drilling parameters recorded in real-time, known as measurement while drilling (MWD). MWD is an instrumented drilling technique where drilling parameters, such as drilling rate, rotational speed, thrust, and torque, are recorded. Despite the advancements in MWD practices in the energy industry, MWD application in geotechnical engineering remains limited because of the lack of comprehensive research utilizing a large database and a systematic analysis of MWD parameters and their relationships with changing stratigraphy. This lack of research motivated this study to perform multiple exploratory drillings using an instrumented drill rig to measure the MWD parameters and calculate the compound parameters. The computed compound parameters were plotted with depth and their predictive capabilities toward soil stratigraphy was assessed. This assessment included an analysis of variation in MWD and compound parameters for both drilling data in the literature and for the data collected in Illinois. Data from all sites were combined and they were used to train various machine learning models, including decision trees, random forests, XGBoost, neural networks, and support vector machines, with hyperparameter tuning. The models were benchmarked for their prediction accuracies. The XGBoost model achieved an accuracy of 0.85, which was the highest compared to other models. The outcomes of this research will contribute to the generation of an extensive dataset for future MWD research and solidify the use of MWD as a cost-effective complementary approach to traditional site investigation techniques.
