A ROBUST HYBRID APPROACH FOR INTRUSION DETECTION IN DYNAMIC AND HETEROGENEOUS IOT ENVIRONMENTS

The rapid growth of the Internet of Things(IoT)has led to the deployment of billions of heterogeneous devices, significantly increasing the attack surface of IoT networks and exposing them to a wide range of cybersecurity threats. In this context, intrusion detection systems (IDS) play a key role in securing these environments. However, existing IDS solutions suffer from several limitations,including reduced generalization capabilities, poor adaptability to dynamic environments, sensitivity to class imbalance, and difficulties in processing massive volumes of heterogeneous data. In this article, we propose a hybrid approach to binary traffic classification (normal and malicious), specifically designed for dynamic IoT environments. This approach combines Auto Encoders (AE),convolutional neural networks (CNN), bidirectional long short-term memory (BiLSTM) neural net-works,and an attention mechanism. The model is based on a complete preprocessing pipeline that integrates feature selection based on five complementary techniques, namely analysis of variance (ANOVA), mutual information (MI), Kendall’s rank correlation, minimum redundancy maximum relevance (mRMR), and the chi-square test. Outliers were handled using winsorization based on the interquartile range (IQR), and class imbalance was corrected by generating synthetic data using AE. The experiments were conducted on the massive NF-ToN-IoT-V2 dataset, comprising more than 13,135,881 observations divided into ten traffic classes.