Residential density classification for sustainable housing development using a machine learning approach

Abstract Using Machine Learning (ML) algorithms for classification of the existing residential neighbourhoods and their spatial characteristics (e.g. density) so as to provide plausible scenarios designing future sustainable housing is a novel application. Here we develop methodology using Random Forests algorithm (in combination with GIS data processing) detect classify within region between Oxford Cambridge, that is, ‘Oxford-Cambridge Arc’. The model based on four pre-defined urban classes, Centre, Urban, Suburban, Rural entire region. resolution grid 500 m × m. features include (1) dwelling geometric attributes garden size, building footprint area, perimeter), (2) street networks length, density, connectivity), (3) density (number units per hectare), (4) types (detached, semi-detached, terraced, flats), (5) surrounding neighbourhoods. results, overall average accuracy 80% (accuracy class: Centre: 38%, Urban 91%, Suburban 83%, 77%), Arc show most important variables were three area: number private gardens. results are used establish baseline current status in bring data-driven decision-making processes level local authority policy makers order support development at regional scale.