Rapid urbanization poses significant challenges to regional sustainable development, particularly in coastal urban agglomerations balancing economic growth and ecological protection. As a vital engine of the Yellow River Basin, the Shandong Peninsula Urban Agglomeration (SDPUA) faces marked spatio-temporal disparities and eco-environmental pressures. Assessing its sustainability and understanding its complex driving mechanisms are essential for achieving the Sustainable Development Goals (SDGs).This study developed an integrated Economic-Social-Environmental (ESE) framework aligned with 17 SDGs. Utilizing multi-source spatial data from 16 prefecture-level cities in SDPUA, we deployed the Entropy Weight Method (EWM) to evaluate regional sustainability scores. Furthermore, a Coupling Coordination Degree Model (CCDM) was applied to analyze spatial synergies, while an explainable Random Forest Algorithm combined with SHapley Additive exPlanations (RFA-SHAP) was established to uncover non-linear driving mechanisms and non-local spatial effects.The results indicate that: (1) SDPUA’s overall sustainability exhibited a spatial pattern of [e.g., higher scores in eastern coastal areas and lower scores in western inland regions], with the economic domain showing the highest internal variance. (2) Coupling coordination across ESE subsystems steadily improved, transitioning from [e.g., moderate coordination to high coordination], though spatial mismatches between economic expansion and environmental capacity persist. (3) The RFA-SHAP analysis identified [e.g., innovation capacity, green cover, and water resource utilization efficiency] as the primary drivers, revealing non-linear threshold effects where key variables significantly boost sustainability only beyond specific critical values.These findings demonstrate the necessity of moving beyond traditional linear assumptions when formulating urban agglomeration policies. To mitigate internal imbalances, regional governance should focus on targeted, threshold-aware interventions and cross-boundary ecological compensation. This study provides a replicable, explainable machine learning framework for SDG-based regional assessment and tailored sustainability policy design.