Journal of Engineering Research and Reports

Machine learning (ML) techniques have become increasingly prominent in power systems applications due to their ability to model complex, nonlinear relationships across large-scale datasets. Existing ML-based studies and survey papers, however, often focus on algorithmic performance comparisons while overlooking critical limitations such as data dependency, generalisation across operating conditions, interpretability, scalability, and practical deployment challenges in real-world power systems. Moreover, many prior reviews treat ML techniques in isolation, with limited emphasis on their suitability for specific power system problems, evolving grid architectures, and operational constraints. This review provides a structured and critical analysis of ML-based approaches applied to power systems, moving beyond descriptive summaries to systematically evaluate their methodological limitations, application gaps, and practical feasibility. Unlike earlier surveys, the paper explicitly categorises ML techniques according to power system domains, highlights unresolved challenges related to data quality, model robustness, and explainability, and examines emerging trends that address these shortcomings. By synthesising recent advances and identifying open research directions, this review offers a comprehensive framework that supports both researchers and practitioners in selecting, adapting, and deploying ML methods for reliable and sustainable power system operation.