Smart Energy Management: A Machine Learning Framework for Predicting Periodic Electricity Demand in a Government Building in Iraq

The surge in need for energy management across utility public infrastructure leads to demands of high-resolution forecasting systems, intelligent in particular third world countries like Iraq that often yield storm clouds littered with intermittent energy supply and unavoidably unpredictable inefficiencies. In this study, we present a smart energy management framework based on Machine Learning (ML) methods which predicts periodic electricity consumption in a government building situated in Baghdad. The proposed technique draws from historical electricity consumption data, meteorological data input and factors within operations: equilibrium between residential and noncommercial usage (all day), public holidays and so on. It is not so with traditional methods. Several machine learning models were developed including Long Short-Term Memory (LSTM), XGBoost, and a Combined Ensemble that integrates these two. The ensemble model achieved the best result in terms of prediction accuracy with a Mean Absolute Percentage Error (MAPE) of 3.79% and an R² score of 0.958. This predictive system can in real time predict electricity demand, supports information-based decision making and load management. Moreover, a practical architecture for scalability of the system was designed together with a smart dashboard allowing visualization and alarms. The findings show how localized ML-based systems could greatly enhance energy efficiency in government buildings, and suggest implications for wider points of energy policy and sustainable development. Future extensions will include a focus on multi-building applications, learning models able to adapt and onwards linking up with sensor systems based on the Internet of Things (IoT).