<p>The continuous casting process plays a pivotal role in modern steel production, directly influencing product quality and economic benefits. In recent years, machine learning has been increasingly applied to address complex and nonlinear metallurgical issues, enabling data-driven prediction, detection, and optimization. The latest progress in applying machine learning to continuous casting was systematically summarized, with a focus on three key areas: abnormal condition prediction, slab quality detection, and process optimization. For abnormal events such as sticking breakout, submerged entry nozzle clogging, and mold level fluctuation, machine learning models exhibit more accurate and adaptive prediction capabilities than traditional threshold-based methods. In defect detection, various defects can be captured by trained models based on computer vision and production process data. In terms of optimization, offline approaches leverage interpretability tools to visualize the decision-making behavior of the model using historical data, whereas online optimization enables real-time decisions and closed-loop control. Importantly, considering the specificity of metallurgical mechanisms, feature selection, model design, and result interpretation need to be guided by domain knowledge, thereby bridging the gap between theoretical algorithms and industrial applicability. This work aims to provide a comprehensive reference for designers, facilitating the development of more efficient and reliable machine learning solutions in the continuous casting process. </p>