A systematic review of analytical machine learning techniques for optimizing decisions in underground engineering

Population growth and rapid urban development have emerged as significant global challenges. These problems are causing high demand for projects constructed below the ground. Although engineers and researchers involved in underground project design are trying to find possible solutions, they still face difficulties in understanding the complex behaviors between underground rocks/soil structures. Machine learning (ML) is a popular hot topic of research that has shown a more remarkable ability to understand complex behaviors of underground engineering systems compared to conventional classical techniques. This paper aims to present a general overview of the progress of ML methods application to address underground issues, specifically in the interaction between soil and structure below ground over the last decade. The review of ML methods focused on aspects such as soil/rock characterization, deep excavations, pile foundations, and tunneling. Additionally, the merits of widely applied methods are presented to assist future researchers in choosing a better approach to implement while studying the same topic. This paper also provides a comparative analysis between traditional underground modeling techniques and emerging ML approaches, exploring how hybrid methods can leverage the strengths of both paradigms. Through detailed case studies, we demonstrate practical implementations while identifying key challenges in deploying ML solutions in field environments. The review includes physics-informed ML frameworks that integrate domain knowledge with data-driven insights, highlighting how these approaches address limitations of purely data-driven models in underground applications. Finally, the shortcomings of ML models discussed in this paper and possible ways of addressing the highlighted problems to improve the prediction performances are also presented.