The proliferation of Internet of Things (IoT) devices has led to massive amounts of sensitive data generation, making data security a paramount concern. Existing methods often struggle with protecting heterogeneous IoT data efficiently, particularly during model training and communication. In this work, we propose a federated learning framework integrated with secure encryption mechanisms to safeguard IoT data during model training and aggregation. Each client device trains a local model using its own sensor data, encrypts the model parameters, and sends them to a server. The server aggregates the encrypted models and sends back the global model for decryption by the clients, ensuring data privacy throughout the process. The proposed framework reduces the unauthorized access risks and also the experimental results demonstrate that the model results in an accuracy of 92% during prediction tasks. The system's encryption overhead was minimal, with only a 7.5% increase in computation time compared to unencrypted federated learning methods. Future work will focus on optimizing the encryption techniques for resource-constrained IoT devices and exploring adaptive security mechanisms powered by machine learning to detect emerging threats dynamically.