In densely populated urban areas, there is an increasing demand towards constructing complex structural systems that incorporate underground facilities situated beneath clusters of aboveground buildings. The dynamic interaction between these subsurface and surface structures, mediated through the surrounding soil, is referred to as structure–soil–structure interaction (SSSI). SSSI is a critical phenomenon that poses significant challenges for the design and analysis of such complex systems. This paper develops an uncertainty-aware machine learning surrogate model, based on a gradient boosting algorithm, to enable the rapid and accurate assessment of SSSI under dynamic loading. The surrogate model is trained using synthetic data sampled employing Latin Hypercube Sampling, which substantially reduces the required dataset size while maintaining high predictive accuracy. The synthetic dataset is produced using a numerical framework that couples the 2.5D Singular Boundary Method for modelling wave propagation in the soil with the Finite Element Method for modelling structural components. To incorporate uncertainty, a dual-ranking strategy is employed that can be flexibly integrated with various surrogate modelling approaches and allows epistemic uncertainty to be embedded directly into the model. The performance of the proposed uncertainty-aware surrogate modelling approach is demonstrated through an application involving three circular structures embedded within a homogeneous full-space. This case study highlights the importance of explicitly accounting for uncertainty and illustrates its influence on predicted output parameters. The results show that the proposed uncertainty-aware surrogate modelling approach is both robust and accurate, outperforming standard surrogate models that neglect uncertainty when applied to SSSI problems.