<p>The rapid development of Internet of Things (IoT) devices and the massive volume of generated data bring into focus the emerging need for an efficient method of distributed data management and aggregation on large scales. Most of the generated data are created at geographically distributed locations; relaying this information to a central server may cause issues related to increased communication costs, processing delays, and a rise in network congestion. One of the new strategies is the use of Federated Learning. With this approach, devices perform local training of their models and share only the model parameters with the central server, thus saving the need to send raw data. Not only does this approach reduce communication costs, but it also significantly enhances data security. However, Federated Learning in IoT environments faces several challenges regarding data heterogeneity across devices, limitations on hardware resources, high communication costs, and the complexity of managing distributed data. This paper proposes a method called FICAD (Federated IoT Clustering and Data Aggregation) for IoT-specific data aggregation and model training. The proposed solution groups data into clusters based on which local models are constructed for each cluster. These aggregated models are further communicated to the central server to generate the final model and minimize communication overhead, making it scalable. Experimental simulation on various datasets confirms that the proposed scheme improved predictive accuracy by 18%, energy efficiency by 15%, and execution time reduction by 11%, showing that the proposed method mitigates IoT challenges.</p>