Hybrid Deep Learning Model for Detecting Malicious Activities in Network Traffic
DOI:
https://doi.org/10.70917/ijcisim-2026-3766Keywords:
Intrusion Detection System (IDS), Deep Learning, Hybrid CNN-BiLSTM, Attention Mechanism, Feature Selection, Class Imbalance, Selective SMOTE, Network Traffic ClassificationAbstract
As cyber threats have evolved, traditional IDS systems struggle to detect malicious network activities. Deep learning has proven to be promising approache for large-scale network traffic analysis, but with redundant features and class imbalance still remaining in the analysis, it creates difficulty in detecting the network trafficMalicious traffic detection using a hybrid deep learning framework combining Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM ) and Attention mechanisms is proposed in this research. CNN extracts spatial traffic features, Bi-LSTM models temporal dependency, and the Attention models relevant traffic patterns. Secondly feature selection is utilized so as to diminish dimensionality and enhance computational economy. A Selective SMOTE strategy is used to boost attack minority classes in order to effectively combat class imbalance. The proposed model was evaluated on large scale network traffic dataset and delivered excellent performance with an accuracy around 97.17% and an ROC-AUC around 99.66.
