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