Electricity Consumption Forecasting Using Time Series Analysis

Forecasting electricity consumption is essential for effective energy management and long-term sustainability. In this study, household electricity demand is forecasted by leveraging historical consumption patterns along with external factors such as temperature, weekends, and holidays. Multiple forecasting models, including ARIMA, SARIMA, and LSTM, were evaluated across both short-term and long-term horizons to explore their strengths and limitations in different scenarios. The dataset used for this research was sourced from the UCI Machine Learning Repository and contains detailed household power measurements such as active power, voltage, and sub-metering for specific appliances. After extensive preprocessing and resampling to ensure data consistency, the models were trained and their performance was compared using evaluation metrics like MAE, RMSE, and MAPE. The results indicate that statistical models such as ARIMA and SARIMA are well suited for short-term forecasts, while deep learning models like LSTM provide better accuracy for long-term predictions. Incorporating external factors through the SARIMAX model further enhanced forecasting accuracy, particularly in capturing spikes in consumption during extreme weather or holiday periods. The findings emphasize the importance of selecting the appropriate forecasting model based on the prediction horizon and integrating contextual data to improve reliability. Additionally, this research highlights the value of hybrid modeling approaches and paves the way for smarter, data-driven strategies that can support efficient energy use and grid stability in increasingly complex consumption environments.