Abstract
Developing accurate predictive models for pile bearing capacity on rock is crucial for optimizing foundation design and ensuring structural stability. This research presents an advanced data-driven framework that integrates multiple machine learning algorithms to predict the bearing capacity of piles based on geotechnical and in-situ test parameters. A comprehensive dataset comprising key influencing factors such as pile dimensions, geological characteristics, and penetration resistance was utilized to train and validate various models, including Kstar, M5Rules, ElasticNet, XNV, and Decision Trees. The Taylor diagram and statistical evaluations demonstrated the superiority of the proposed models in capturing complex nonlinear relationships, with high correlation coefficients and low root mean square errors indicating robust predictive capabilities. Sensitivity analyses using Hoffman and Gardener's approach and SHAP values identified the most influential parameters, revealing that penetration resistance, pile embedment depth, and geological conditions significantly impact pile capacity. The findings underscore the effectiveness of machine learning in geotechnical engineering applications, offering a reliable and efficient alternative to traditional empirical and analytical methods. The developed framework provides engineers and practitioners with a powerful tool for improving pile design accuracy, reducing uncertainties, and optimizing construction practices. Future research should focus on expanding the dataset with diverse geological conditions and exploring hybrid modeling techniques to enhance prediction accuracy further.