Using machine learning to examine associations between the built environment and physical function: A feasibility study