Real-Time Intrusion Identification in IoT Networks Based on Hybrid Deep Learning Networks
DOI:
https://doi.org/10.30750/Keywords:
Cyberattack, Intrusion Detection, IoT, Deep Learning, N-BaIoT dataset, CNN+LSTM.Abstract
New vulnerabilities have emerged as a result of the fast expansion of IoT networks, calling for the installation of advanced intrusion detection systems
(IDS) that can identify complex attack patterns. The authors of this work propose a CNN-BiLSTM deep learning model to enhance detection accuracy
in Internet of Things (IoT) settings. The model extracts geographical and temporal features from the N-BaIoT dataset to classify network behavior.
The suggested model beats the competition with a 98.67% F1-score (F1), 99.91% recall (REC), 98.56% precision (PRE), and 99.79% accuracy
(ACC), according to the testing results. The confusion matrix confirms excellent accuracy in categorizing all classes, and there is little overfitting
and speedy convergence in the training and testing curves. Comparative analysis with traditional models such as Multilayer Perceptron (MLP),
AdaBoost, and Naive Bayes (NB) confirms the robustness and scalability of the CNN-BiLSTM architecture. The results emphasize its capability to
be implemented in real-time in the dynamic IoT ecosystems as a credible and smart approach to next-generation network security
