The integration of renewable energy sources and the increasing complexity of power grids has led to the development of smart grids that utilize advanced sensing, communication, and control technologies. This review presents a critical discussion on the current advances in ma- chine learning (ML) techniques for load balancing and fault detection in smart grids. Frameworks for smart grid management are described, and essential sub-disciplines are reviewed, including data-intensive ML strategies and algorithms, smart grid databases and data generation approaches, and key power system descriptors used in ML models. Furthermore, an in-depth discussion on the applications of ML in load forecasting, demand response, fault detection, and self-healing is provided. Finally, how these sub-disciplines support data-driven smart grid management is outlined, and the opportunities and challenges are highlighted.