Machine Learning Prediction Tool for Seismic Bearing Capacity of Strip Footings in Rock Mass

This study evaluates the potential of various machine learning (ML) models in predicting the seismic bearing capacity of strip footings in rock mass. Six ML algorithms, i.e., multilinear regression (MLR), K-nearest neighbor (KNN), support vector machine (SVM), decision trees (DT), random forest (RF), and extreme gradient boosting (XGBoost) were adopted. A database comprising 960 samples based on the results of robust finite element limit analysis was employed for the purpose of training and testing. The factors considered in the database include the rock mass parameters, such as the geological strength index (GSI), rock yield parameter (mi), unconfined compressive strength of the intact rock (σci), density of the rock (γ), width of the strip footing (B), depth of embedment (d), and the horizontal seismic coefficient (kh). The input parameters considered to train the ML models included the GSI, mi, kh, rock strength ratio (σci/γB), and embedment depth ratio (d/B) while the seismic bearing capacity factor N, which is the ratio of the ultimate seismic bearing capacity (qu) to the unconfined compressive strength of the rock (σci) was taken as the output. The performance of the trained ML models was evaluated using performance metrics, such as R-squared, root mean squared error (RMSE), and mean squared error (MSE). The results revealed that XGboost shows the best performance (R2=0.999) in comparison to other ML algorithms. Considering the XGBoost model, further analysis was performed to assess the relative importance of the parameters on the output. GSI was found to be the most influential parameter followed by mi. The trained XGBoost model was used to develop a web application that can be used to determine the seismic bearing capacity of strip footing in rock mass.