This chapter explores advanced applications of machine learning techniques to enhance the security of networks-on-chip (NoCs). While previous chapters concentrated on CNN-based countermeasures for detecting abnormal data in NoCs, this chapter expands the discussion to highlight how diverse machine learning models can address a broader range of security challenges. By introducing two representative schemes, namely, cascaded machine learning models for mitigating general DoS attacks and a graph-based early defense mechanism against refined flooding DoS attacks, the chapter demonstrates the advantages of combining multiple models. These approaches strategically utilize different machine learning techniques across various stages of data processing to improve detection accuracy and ensure timely responses to security threats. Additionally, the chapter discusses a case study of a novel secure NoC microarchitecture that incorporates advanced time-series analysis models such as GNNs and LSTMs. This innovative architecture enhances real-time anomaly detection capabilities and supports adaptive responses to dynamic security threats while maintaining resource efficiency. The chapter provides a comprehensive overview of threat modelling, detailed descriptions of the defense mechanisms, and experimental evaluations to validate their effectiveness. By bridging advancements in machine learning with hardware security, this chapter establishes a foundation for further innovation in safeguarding large-scale NoCs.