Exploring Artificial Intelligence Models for Rock Mass Classification: An Assessment of Cost Estimation

Abstract

This thesis investigates the application of machine learning (ML) models to enhance rock mass classification systems in geotechnical engineering. Traditional classification methods, such as the Rock Mass Rating (RMR) and Geological Strength Index (GSI), have been widely used for decades. However, these approaches often rely on subjective human assessment and are limited in handling complex geological conditions and dynamic environments. This research explores the potential of artificial intelligence (AI) and machine learning to address these limitations and improve rock mass classification accuracy, consistency, and adaptability.
Using a comprehensive borehole dataset with key geological features such as Rock Quality Designation (RQD), Joint Roughness Number (Jr), and Stress Reduction Factor (SRF), various machine learning models were developed and evaluated. A logistic regression model was employed as the primary AI-based approach, achieving a high classification accuracy of 98.03%, significantly improving over the baseline dummy classifier's accuracy of 24.29%. The study demonstrates that machine learning models can dramatically reduce the subjectivity associated with traditional methods by relying on data-driven insights and advanced statistical techniques.
The research also highlights the critical role of data preprocessing, feature selection, and model evaluation metrics in optimizing machine learning models for geotechnical applications. Visual tools such as heatmaps and confusion matrices were used to analyze model performance and identify areas for improvement. The findings emphasize the importance of high-quality data and suggest that when properly trained and validated, AI models can effectively classify rock masses, supporting better decision- making in construction, tunnelling, and mining projects.
Despite the promising results, the study acknowledges challenges related to data quality, model complexity, and the need for further research on integrating AI systems into practical geotechnical workflows. The thesis concludes that machine learning offers a transformative approach to rock mass classification, providing significant advancements in accuracy and efficiency over traditional methods.
This research provides a foundation for future studies aimed at developing more robust AI-driven geotechnical solutions, which would contribute to safer and more efficient engineering projects.