DOI: 10.35377/saucis...1777102 ISSN: 2636-8129

Deep Learning Approaches for Intrusion Detection System Using Multi-Model Architectures for Network Security

Yaseen Yaseen, Aythem Kareem, Mohammed M. Al-ani, Ahmed Adil Nafea
Show PDF Cite
It has become necessary to design efficient, robust, and effective Intrusion Detection Systems (IDS) amid the rising complexity of cyberattacks and the exponential growth in network traffic. In this work, we present an advanced IDS framework using comparative deep learning (in short, DL) models for precise classification and detection of all types of network intrusions. The proposed system consists of five DL models: Recurrent Neural Network (RNN), One-Dimensional Convolutional Neural Network (1DCNN), Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), and a Transformer-based model, which are trained and tested on the CICIDS2017 and CICIDS2018 datasets, whose network characteristics are close to real-world network traces. The Input features were pre-processed with data encoding and standardization to facilitate and make them compatible with optimal learning for models. The dataset was split into a 70% training set and a 30% test set to evaluate the generalization of the models. We used F1 score, recall, precision, and accuracy as metrics to evaluate the performance. The convolutional models, especially the 1DCNN, perform better due to the feature extraction using convolutions, achieving 99.20%, while LSTM achieved 98.68% and RNN 98.53%. These findings are further indicative of the potential that DL-based hybrid frameworks present for developing an intelligent, scalable, and time-sensitive IDS enabling practical applicability.

More from our Archive