<p>Smart Home Energy Management Systems (SHEMS) play a vital role in improving residential energy efficiency, reducing costs, and supporting the integration of renewable energy sources. This review systematically examines recent advancements in Machine Learning (ML) and Deep Learning (DL) techniques for optimizing SHEMS, covering studies published between 2018 and 2024. Using the PRISMA-based structured review methodology, 80 high-quality research articles are analysed to evaluate algorithmic performance, technical challenges, security considerations, and economic viability. The findings indicate that traditional ML models, such as Support Vector Machines (SVM), and regression techniques, remain effective for structured data and resource-constrained environments. However, DL approaches, including Long Short-Term Memory (LSTM), CNN-based hybrid models, and Deep Reinforcement Learning, consistently outperform conventional methods in real-time energy demand forecasting, adaptive load scheduling, and handling renewable energy intermittency, achieving prediction accuracies above 95% and energy savings of 25–40%. Hybrid models integrating ML/DL with metaheuristic optimization techniques further enhance system robustness and multi-objective optimization performance. The review also highlights growing attention to security and privacy, where advanced encryption and multi-factor authentication significantly improve protection, though implementation gaps remain. Economic analysis reveals that intelligent optimization improves long-term return on investment despite high initial costs. Overall, the study identifies key research gaps in benchmarking, scalability, security validation, and techno-economic assessment, providing directions for future intelligent and secure SHEMS deployment.</p>