Title: Supervised Deep Learning for Anomaly and Intrusion Detection in Big IoT Data: A Review of Techniques and Open Challenges
Authors: Virendra Tank
Volume: 10
Issue: 7
Pages: 399-407
Publication Date: 2026/07/28
Abstract:
The proliferation of the Internet of Things (IoT) has produced an unprecedented volume, velocity, and variety of network traffic, commonly characterized as "Big IoT Data." This scale, combined with the heterogeneity and resource constraints of IoT devices, has rendered conventional signature-based and shallow machine-learning intrusion detection systems (IDS) inadequate against sophisticated and evolving cyber-attacks. Supervised deep learning (DL) has emerged as a dominant paradigm for anomaly and intrusion detection because it can automatically learn hierarchical spatial and temporal representations directly from raw or lightly processed traffic features. This review synthesizes 26 peer-reviewed and preprint studies published between 1997 and 2025 to examine how supervised DL architectures - including Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks, Deep Belief Networks (DBN), hybrid CNN-LSTM/CNN-GRU models, and attention/transformer-based models - have been applied to big IoT intrusion and anomaly detection. A structured methodology, benchmark datasets (NSL-KDD, UNSW-NB15, CICIDS2017, BoT-IoT, TON_IoT and IoTID20), and evaluation metrics are discussed, followed by a comparative analysis of reported detection performance. The review identifies open challenges related to class imbalance, adversarial robustness, real-time inference on constrained devices, explainability, and dataset representativeness, and it outlines future research directions including federated and edge-native learning, self-supervised pretraining, and lightweight transformer variants for resource-constrained IoT environments.