Application of Machine Learning Techniques to Predict Unconfined Compressive Strength of Sedimentary Rocks in UAE

The unconfined compressive strength (UCS) is a crucial factor in geotechnical engineering analysis and design, despite the strict test limitations set by codes. In order to ease the UCS prediction with physical testing, extensive research has been conducted. However, two major challenges in predicting UCS for sedimentary rocks in coastal regions of the United Arab Emirates (UAE) are the weathering conditions of the rocks and the cost of extracting and performing the physical tests. To overcome these obstacles, data was collected from both field and laboratory tests, including the depth of the sample, type of rocks, and location in the UAE, as well as point load, ultra-sonic pulse velocity, and UCS tests. Using MATLAB, multiple Machine Learning regression models were developed to predict UCS for these sedimentary rocks. Of these models, the linear Support Vector Machine performed the best, with a high estimated R2 value of 0.91, RMSE of 1.43, and MAE of 0.89 for training. In comparison, the testing phase yielded an R2 value of 0.75, RMSE of 2.31, and MAE of 1.47. These results suggest that Machine Learning models can effectively predict UCS for sedimentary rocks in coastal regions of the UAE, despite the challenges posed by weathering conditions and cost constraints.