This study proposes a spatial machine learning model that integrates the core principles of the random forest model with the spatial partitioning rules of geographically weighted regression to investigate the spatial nonlinear effects of built environment factors on dockless bikesharing (DBS)–metro integrated use. Using DBS data from Beijing, China, four types of access and egress integrated use during peak hours were calculated through an optimized spatial network density-based method. Twenty-one independent variables related to socioeconomic and demographic, land use, road network design, transport facility, and station attributes, were analyzed based on varying catchment area sizes. The results demonstrate that all the built environment variables exhibit nonlinear effects on integrated use, with most of their effective ranges and threshold effects varying among stations. Population density and housing prices emerge as dominant variables affecting morning access integrated use. The positive impact of commercial places is more pronounced in suburban regions, while the influence of educational places is stronger in the northern city. Stations located at a moderate distance to city center (DC) account for the majority of integrated use trips, with DC exerting a greater influence on suburban areas than on urban areas. These findings offer valuable insights for enhancing seamless connections between bikesharing and metro systems.