In the progressing realm of the Internet of Things (IoT), the maintenance of strong security is identified as a key challenge. Conventional intrusion detection systems (IDS) frequently prove inadequate considering the dynamically distributed characteristics of IoT settings. The current study introduces a specialized intrusion detection system for IoT networks, incorporating adaptive machine learning techniques. By leveraging advanced techniques in federated and split learning, our approach dynamically adjusts to the evolving threat landscape, ensuring high detection accuracy while preserving data privacy. We investigate the incorporation of lightweight machine learning models for the purpose of supporting resource-limited IoT devices, presenting a thorough examination of their effectiveness. Empirical findings suggest that our adaptable Intrusion Detection System (IDS) not only boosts detection capabilities but also minimizes false alarms, presenting a scalable and efficient technique for securing IoT environments. This work underscores the potential of adaptive machine learning methodologies in fortifying IoT networks against sophisticated cyber threats.