Ground subsidence resulting from underground coal mining poses significant challenges to urban safety, infrastructure stability, and environmental protection, particularly in regions extending beneath water bodies. This study investigates subsidence patterns in the Kozlu coal basin by integrating Interferometric Synthetic Aperture Radar (InSAR), numerical modelling, and machine learning techniques. The Kozlu coal basin, located in Zonguldak, Turkey, serves as a critical example, where extensive mining activities have led to complex deformation patterns. InSAR effectively captures terrestrial subsidence but is limited in underwater regions. Numerical modelling provides insights into geological behaviour but requires extensive input data. Machine learning, specifically Gaussian Process Regression (GPR), bridges this gap by predicting subsidence in unobservable underwater zones with high accuracy. The integrated approach reveals consistent deformation trends across terrestrial and marine environments, offering practical tools for risk mitigation and resource management. These findings underscore the importance of interdisciplinary methods in addressing complex geological challenges and pave the way for future advancements in subsidence monitoring and prediction.