An Intelligent Cybersecurity and Privacy‐Preserving Internet of Drones Framework Based on Federated Learning and Hybrid Deep Ensemble Models

The advent of drone‐based novel cybersecurity applications combined with the Internet of Things (IoT) has transformed navigational technologies with effectual data communication services on numerous platforms. This progression employs machine learning (ML) and deep learning (DL) techniques for future advancement. Currently, there has been a considerable rise in the employment of IoT‐based drone data management system. From traditional sectors like heavy industry to advancements in precision agriculture, a trend is mirrored by the integration of smart cities for smart and effective observation. Numerous organizations are not aware of the huge amount of IoT devices and how they might pose cybersecurity and privacy hazards. With data confidentiality and integrity becoming a main concern recently, a novel technology like federated learning (FL) has been developed. FL is a current growth in artificial intelligence (AI) that normally depends on the decentralized data model. FL has obtained many applications in cybersecurity regions. This paper develops a Federated Learning for Intelligent Cybersecurity and Privacy Preservation Using Deep Ensemble Models (FLICPP‐DEM) approach in Internet of Drones. The FLICPP‐DEM system focuses on leveraging FL to improve cybersecurity in drone‐enabled IoT ecosystems while preventing real‐time cyberattacks. At first, the min–max normalization was applied in the data preprocessing phase for cleaning, transforming, and organizing raw input data into a structured form. In addition, the recursive feature elimination (RFE) as well as minimum redundancy maximum relevance (MRMR) models are applied for the feature selection procedure to select and identify related features from the dataset. Then, the presented FLICPP‐DEM methodology deploys an ensemble model, namely, the deep belief networks (DBN) method, the temporal convolutional networks (TCN) model, and the variational autoencoder (VAE) technique for the cybersecurity classification process. A wide range of simulations was included to show the efficacious performances of the FLICPP‐DEM method. The stimulated results inferred the proficient performance of the FLICPP‐DEM system in the cybersecurity process of classification over the existing approaches. The proposed FLICPP‐DEM model achieves an accuracy of 99.65%, demonstrating its effectiveness. The results indicate improved performance and robustness over existing methods.