A Hybrid Ensemble Framework for IoT Network Anomaly Detection Using Advanced Autoencoders and Specialist Learners
Open Access DepositedThe exponential growth of Internet of Things (IoT) devices has introduced substantial cybersecurity challenges, particularly in the detection of low-frequency but high-impact network anomalies. Traditional anomaly detection systems often fail to generalize across highly imbalanced datasets and struggle with the high dimensionality characteristic of IoT network traffic. This praxis introduces a hybrid ensemble framework centered around the Network Feature Variational Autoencoder (NFVAE), a custom architecture that integrates multiscale feature extraction, attention-based fusion of latent and reconstruction pathways, learnable feature importance weighting, and class-specific specialist branches. The NFVAE is embedded within a calibrated ensemble comprising Random Forest and XGBoost classifiers, each tuned to enhance detection of underrepresented attack classes. Evaluated on the CICIoT2023 dataset, the proposed framework achieved a .49% FPR in binary classification and recall improvements up to 66% for individual attack classes, with an average recall increase of 27% for minority classes in the multiclass setting, contributing directly to an increase in macro recall and macro F1-score. These findings advance the state of IoT anomaly detection and provide a reproducible framework for building scalable, adaptive, and resilient intrusion detection systems in real-world environments.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.