Remmel, TarmoRussell, William2026-11-052026-11-052026-08-31Major Paper, Master of Environmental Studies, Faculty of Environmental and Urban Change, York Universityhttps://hdl.handle.net/10315/44136Housing affordability has become a notorious planning challenge in Canada, particularly within the Greater Toronto-Hamilton Area (GTHA), where rising housing costs have exacerbated inequalities. While existing research has extensively documented the political economy of housing and the consequences of neoliberal housing policy, fewer studies have applied machine learning techniques to identify the relative predictive power of socioeconomic, housing, accessibility, and land-use variables in explaining census tract-level affordability. This study addresses this gap through a spatial analysis of 2021 census tract data for the GTHA. Using Geographic Information Systems (GIS), approximately 150 variables from Statistics Canada, the Canada Mortgage and Housing Corporation, municipal open data, and provincial datasets were compiled and processed into a unified spatial database. Classification and Regression Trees (CART) were used to explore the predictive relationships for housing affordability outcomes among variables before developing Random Forest regression models to predict the proportion of households living in affordable housing, defined using the Canada Mortgage and Housing Corporation's shelter-cost-to-income ratio (STIR) of below 30%. Getis-Ord Gi* statistics were subsequently applied to the residuals of the final model to identify statistically significant spatial clusters where affordability outcomes remained unexplained. The final Random Forest model explained approximately 79.7% of the variation in housing affordability across census tracts, demonstrating that housing affordability is strongly associated with interacting socioeconomic and housing-market characteristics rather than any single determinant. Variables relating to housing tenure, household composition, immigrant status, housing costs, subsidized housing, and building age emerged as the strongest predictors, whereas older housing builds and housing suitability categories contributed little to no predictive value. Residual hotspot analysis further revealed localized areas where unknown factors influence affordability beyond those captured by the model. By integrating critical urban planning theory with explainable machine-learning methods, this research demonstrates how spatial data science can support planning practice while identifying structural patterns of housing inequality that warrant further policy attention.enHousingThe statePolicyEquity planningResearch Paper