The transition toward sustainable household energy management has become a pressing priority due to rising electricity costs, climate change concerns, and the proliferation of digital appliances. Traditional monitoring techniques often fail to capture dynamic, time-varying patterns of household consumption, limiting their ability to support optimization and demand forecasting. Recent advances in machine learning (ML) offer scalable tools to analyze granular energy data, identify influential consumption drivers, and generate accurate forecasts for demand-side management. This study develops a predictive framework using the UCI Household Power Consumption dataset (2006–2010), integrating exploratory data analysis (EDA), feature engineering, multicollinearity checks, and regression modeling. The results show that Stochastic Gradient Descent (SGD) regression achieves higher scalability and performance (R2 = 0.9520, RMSE = 0.236) while Ordinary Least Squares (OLS) regression offers a good baseline (R2 = 0.9477). The disproportionate contribution of appliances like air conditioners and water heaters to peak loads is further highlighted by sub-metering study. The results highlight the potential of lightweight machine learning models for real-time forecasting in environments with limited resources, providing useful information for utilities, consumers, and legislators to improve energy efficiency and support sustainability objectives.