The evolution of Radio Access Networks (RAN), particularly with the advent of 5G and Open RAN (O-RAN) architectures, has introduced unprecedented complexity and operational demands. Traditional fault management approaches—relying on static thresholds and manual diagnostics—are increasingly inadequate in addressing the dynamic and heterogeneous nature of modern RAN environments. This chapter presents a comprehensive framework for integrating machine learning (ML) into RAN operations to enable intelligent anomaly detection and predictive maintenance. It explores the types of anomalies affecting RAN performance, including KPI degradations, hardware faults, software crashes, and cross-domain failures. A range of ML techniques—spanning supervised, unsupervised, and time-series models—are discussed in the context of their applicability to high-dimensional telemetry data. The proposed system architecture leverages modular, cloud-native components and interfaces with O-RAN elements such as the RAN Intelligent Controller (RIC) and Service Management and Orchestration (SMO) platforms. Real-world use cases demonstrate how ML models can proactively detect faults, reduce downtime, and optimize operational efficiency. The chapter concludes with future directions, highlighting emerging trends such as federated learning, explainable AI, and edge intelligence, which promise to further enhance the resilience and autonomy of next-generation mobile networks.