Socio-Demographic insights on urban building energy consumption

• Socio-demographic influences on energy use: Income, employment, age distribution, household size, and housing characteristics significantly shape both long-term heating behavior and short-term electricity load patterns. • Implications for demand-side management (DSM) and equity: Energy programs should incorporate socio-demographic insights to improve energy savings, targeting behavioral interventions, retrofits, and measures that enhance thermal efficiency. • Methodological contribution: Combines change-point analysis, double Machine Learning, and SHAP-based interpretability to quantify socio-behavioral impacts on residential energy consumption. Provides a multi-dimensional framework integrating long-term structural and short-term behavioral analysis for urban energy demand modeling. Residential buildings are central to global decarbonization efforts, accounting for nearly 40 % of energy use and around 30 % of greenhouse gas emissions. While technological improvements–such as efficient appliances, retrofits, and demand-side management (DSM) programs–are widely promoted, heterogeneity in occupant behaviour remains a major source of consumption variability. Socio-demographic characteristics shape these behaviours, yet their influence on residential electricity use at urban scales remains insufficiently quantified. This study examines how socio-demographic factors drive long-term and short-term electricity consumption patterns across urban neighbourhoods in Québec, Canada. Using hourly electricity data linked with census-derived socio-demographic variables, the analysis proceeds in two phases. First, long-term heating-related behaviour is modelled through change-point analysis, and its associations with socio-demographic attributes are quantified using XGBoost regression with SHAP interpretation. Second, short-term behavioural patterns are assessed by clustering daily load profiles and predicting cluster membership through an XGBoost classification model, again interpreted using SHAP. Results show that income, household size, age structure, employment levels, and housing characteristics significantly influence both structural heating behaviour and daily load shapes. Higher-income and more densely occupied areas display elevated base loads and steeper heating slopes, while lower-income neighbourhoods exhibit earlier heating activation, suggesting reduced efficiency or comfort constraints. Daily load variability is strongly shaped by mobility and work routines: car-dependent and full-time working populations show pronounced morning and evening peaks, whereas neighbourhoods with higher unemployment or walkability exhibit flatter profiles. Aging populations and smaller households sustain higher per-capita use, likely due to longer occupancy durations and comfort-driven heating practices. These findings demonstrate that electricity demand is shaped not only by building attributes but by socio-economic realities and behavioural norms. Embedding socio-demographic diversity into urban energy models provides actionable insights for targeted retrofits, behaviour-sensitive DSM initiatives, and equitable energy policy design.