Highly-priced properties cause affordability problems among low and middle-income buyers. To overcome this, the Malaysian government introduces affordable housing through National Urbanization Policy, National Physical Plan, National Housing Policy, and Eleventh Malaysia Plan. Whilst having a good market response, some areas experience either shortage or surplus of houses reflecting ineffective affordable housing policies. Inappropriate estimation techniques and aggregate location estimations limit the accuracy and usability of demand estimations. Thus, this research aims to establish a framework to estimate local demands for affordable housing. This study selects and reviews the theoretical and modelling framework of the Artificial Neural Network Model (ANN) due to its superior performance in forecasting demand. The ANN theoretical and modelling framework guides the modelling process, which includes data collection and preparation, model development, data analysis and model evaluation. Potential sites for affordable housing development identified from the model’s coefficients are visualised spatially through Geographic Information System (GIS). Localised housing demand forecasts are highly beneficial for policy-makers and housing developers to allocate the number of supplies across locations. This allows maximum take-up rate for affordable housing, avoids supply and demand mismatch and thus achieves the national housing policy agenda.