Cloud-edge environments are increasingly targeted by cyber threats, requiring solutions that ensure both high detection precision and robust data privacy. This paper proposes a privacy-preserving Intrusion Detection System (IDS) by integrating Federated Learning (FL) and Docker-based edge simulations. Rather than transmitting raw network data to centralized servers, our approach trains deep learning models locally within Docker containers emulating edge devices, ensuring sensitive data remains decentralized. The framework employs the NSL-KDD dataset and applies preprocessing techniques such as one hot encoding, min-max normalization, and class balancing via RandomOverSampler. Each local model is optimized using Keras Tuner, and model updates are aggregated using the Federated Averaging algorithm to build a global model. This method not only preserves user data confidentiality but also ensures scalability and adaptability across heterogeneous network environments. We achieve a detection accuracy of 99.78%, with significant improvements in identifying rare attacks such as Remote-to-Local (R2L) and User-to-Root (U2R). While primarily designed for simulation, the architecture is modular and deployment ready, making it suitable for real-world IoT and cloud-edge infrastructures.