FSTs are advanced solid-state processing methods that address the growing industrial demand for lightweight components with enhanced mechanical properties. These techniques, including friction stir welding and friction stir processing, are distinguished by their capability to refine microstructures and improve the quality and longevity of welds and surfaces, making them integral to modern manufacturing. Recent advancements in machine learning (ML) have facilitated the integration of data-driven approaches into FST applications, demonstrating significant potential for optimising performance. This review explores the use of ML in FSTs, highlighting how various ML models improve the prediction of mechanical properties and the optimisation of processing parameters. Findings indicate that ML provides higher accuracy in predictions for FST applications than traditional statistical methods, while hybrid ML techniques further enhance outcomes by refining process control. The review further highlights existing knowledge gaps and proposes directions for future research to enhance ML integration in FSTs. This comprehensive synthesis is drawn from academic literature primarily sourced from the Scopus and Web of Science databases, supplemented by insights from recent books published in the past 15 years.