A Deep Learning-based Generalization Model for Intrusion Detection Across Multiple Data Sources

Abstract

Significantly, the amount of data transferred through the existing communication systems has shown an increase in recent years. Intrusion Detection in Networks: It is observed that the network infrastructures developed in recent years need to be protected from cyber attacks using intrusion detection in networks. Because deep learning intrusion detection systems can recognize intricate patterns in networks, it is crucial to integrate them into networks. Conventional intrusion detection systems that use signature detection techniques are found to be ineffective in identifying unknown types of cyber attacks.This paper proposes an Analytical Comparison of various Cognitive Neural Modeling and Fusion-Based Deep Learning Architectures for Network Intrusion Detection using benchmark and real-time datasets. The proposed IDS system’s capacity for generalization is validated through experiments on the NSL-KDD dataset benchmark Labeled Data Corpus, and the outcomes are compared with those of the most advanced IDS systems. In addition, the real-time dataset is generated using the experimentally controlled environment with two laptops. The real-time dataset is based on both regular network communication and network communication under various attack scenarios.The effectiveness of various models integrating DNN and CNN frameworks for hierarchical and spatial feature extraction, Bidirectional LSTM, TabNet, Hybrid models in identifying. Empirical comparison with next-generation deep learning architectures neural models reveals that the integrated CNN-driven hybrid framework-BiLSTM Attention model attains peak performance in terms of detection process prediction effectiveness of about 96%.The effectiveness and usability of the suggested approach in the creation of modern intrusion detection systems are confirmed by a comparative analysis of the method using the real-time dataset and the NSL-KDD benchmark dataset.