Accurate short-term electricity load forecasting is essential for reliable operation of modern power systems and smart grid infrastructures. This study investigates the temporal dependency structure of household electricity consumption using supervised machine learning techniques. The forecasting task is formulated as a regression problem based on lagged load values, calendar-related temporal features, and exogenous meteorological variables, including air temperature, relative humidity, precipitation, and wind speed. Linear Regression is employed as a benchmark model, while Random Forest is used as a nonlinear ensemble approach. Model performance is evaluated using Mean Absolute Error, Root Mean Square Error, and Mean Absolute Percentage Error under chronological train-test splitting in order to preserve temporal causality. In addition, lag-window sensitivity analysis, walk-forward validation, multi-step forecasting horizon analysis, and feature importance assessment are conducted. The Random Forest model with weather variables achieved the best one-hour-ahead forecasting performance, with MAE of 0.397, RMSE of 0.552, and MAPE of 54.27%. The results show that recent load history and daily periodicity remain the dominant predictors, while meteorological variables provide a limited but measurable contribution. The findings support the use of ensemble learning for short-term load forecasting analysis and highlight the need for further refinement when applying such models to highly volatile residential consumption patterns.