Software-defined networks (SDN) have revolutionized network management by enabling a separation of the control plane and the data plane, but this poses challenges in routing optimization. This paper examines the integration of machine learning (ML) techniques to improve routing in SDN, exploring supervised, unsupervised, and reinforcement learning. We analyze the advantages and limitations of these approaches in key areas such as congestion management, slowness, packet loss, and quality of service (QoS). We also discuss the challenges, including computational complexity, model scalability, and real-time adaptability. This review aims to provide a perspective on the current state of ML application in SDN routing while highlighting the importance of these techniques to improve network performance. Finally, we propose avenues for future research, such as developing hybrid models and creating efficient algorithms for real-time processing. This work highlights the potential of ML to transform SDNs, while identifying barriers to widespread adoption.