<p>This study aims to elucidate the specific mechanisms by which concentrated rainwater infiltration from bioretention facilities affects the deformation of subgrade slopes in mountainous urban sponge roads, with a particular focus on the critical role of facility structural dimensions. Three-dimensional finite element numerical simulation was employed to comparatively investigate the deformation patterns of the subgrade slope under varying bioretention facility dimensions. In addition, a prediction model of sponge road slope deformation based on BP neural network is further constructed, and the model is verified by using actual engineering monitoring data. The results demonstrate that increasing both the width and depth of bioretention facilities significantly amplifies the horizontal displacement at the slope toe and the vertical displacement within the subgrade, with the influence of width being more pronounced than that of depth. The predictions generated by the BP neural network model exhibited good agreement with the actual monitoring values, indicating high predictive accuracy and computational efficiency. This research systematically quantified the quantitative relationships between bioretention facility dimensions and slope displacement in mountainous sponge roads. By innovatively integrating numerical simulation with machine learning methodologies, a high-precision slope deformation prediction model was established. This work provides a direct theoretical foundation and technical support for scientifically determining safe size thresholds for bioretention facilities and for enhancing the predictive capability and risk management of slope stability in sponge city road infrastructure.</p>