A method is provided to using a machine learning model to predict lane geometry where incorrect or missing lane line geometry is detected. Methods may include: receiving a representation of lane line geometry for one or more roads of a road network; identifying an area within the representation as broken lane line geometry of an intersection using a machine learning model; generating a masked area of the broken lane line geometry of the intersection within the representation; processing the representation with the masked area using an inpainting model to generate an inpainted result within the masked area of restored lane line geometry of the intersection, where the inpainting model is trained using a set of representations identified as lane line geometry of intersections; and updating a map database to include the restored lane line geometry of the intersection in place of the broken lane line geometry of the intersection.