With the advent of 5G networks and the expansion of the Internet of Things (IoT), the increase in connected devices has intensified the risk of Distributed Denial of Service (DDoS) attacks with the need for privacy preservation. Recently, Software-Defined Networks (SDN) with P4-based programmable data planes have emerged as an effective solution to these threats due to features like advanced packet processing and the ability to integrate with machine learning. This paper proposes a network-assisted solution combining a programmable data plane with federated distributed learning to detect DDoS attacks using Multi-Layer Perceptron (MLP) to ensure privacy in IoT environments. This approach shifts the preprocessing stage to network switches, enabling real-time traffic feature extraction. In addition to reduce latency and to improve preprocessing accuracy, it also decreases the processing load on local federated servers. To evaluate this solution, three packet processing methods, packet copy, header copy, and metadata extraction, are tested, with metadata extraction achieving a classification time of less than 300 ms. MLP showed good performance with an F1 score above 98%. By reducing latency, enhancing accuracy, and preserving privacy crucial in advanced networks like 5G. This approach serves as an effective solution for DDoS detection in IoT.