An Artificial Intelligence-Driven Predictive Framework for Residential Energy Demand Forecasting in Smart Urban Environments

Reliable short-term electricity demand forecasting is a critical requirement for the operational efficiency and sustainability of smart city energy infrastructure.Insufficient forecast precision leads to either excess generation capacity, causing fuel wastage and elevated carbon emissions, or supply shortfalls that threaten grid stability.This paper presents a systematic comparison of five forecasting approaches applied to residential electricity demand prediction: Long Short-Term Memory (LSTM), Gated Recurrent Unit, Convolutional Neural Network combined with LSTM, Prophet, and eXtreme Gradient Boosting (XGBoost).Experiments were conducted on the UCI (University of California, Irvine) Individual Household Electric Power Consumption dataset, which contains minute-resolution measurements of electricity usage across eight electrical attributes spanning approximately four years.A chronological 80/20 train-test split was applied throughout to ensure that model evaluation reflected performance on genuinely unseen future data.XGBoost achieved the highest predictive accuracy among the five models, attaining a root mean squared error of 0.024 kWh and a normalised root mean squared error of 2.37%, with substantially lower computational cost than the neural network approaches.Prophet captured seasonal demand patterns effectively and yielded interpretable forecast components suitable for integration with real-time urban energy monitoring systems.Among the deep learning models, Convolutional Neural Network-LSTM outperformed both standalone recurrent architectures, confirming the advantage of combining local feature extraction with sequential modelling.The results demonstrate that gradient-boosted tree ensembles, when combined with well-designed feature engineering, constitute a computationally efficient and highly accurate alternative to deep sequence models for short-term residential energy forecasting and are well-suited to deployment within smart city energy management systems.