In modern computer networks, optimizing routing decisions is critical for main- taining high performance, low latency, and efficient resource utilization. The programmability offered by Software-Defined Networks (SDNs) has paved the path for the adoption of Machine Learning (ML) solutions to create innovative routing schemas. In particular, this thesis integrates Reinforcement Learning (RL) and Federated Learning (FL) within SDN to create a global routing model trained in a cooperative fashion. Reinforcement Learning autonomously learns optimal routing policies that adapt to changing network conditions. Concurrently, Federated Learning allows multiple networks to collaboratively develop a global model without sharing raw traffic data, thereby maintaining privacy.