ENHANCING NETWORK SECURITY IN ANOMALY-BASED INTRUSION DETECTION SYSTEMS THROUGH ADAPTIVE NEURAL NETWORKS

Abstract

ABSTRACT While there is continuous change in the world of cyber threats, conventional methods of network security are faltering behind in comparison with sophistication and novelty introduced by some attacks. This research aims to enhance network security by embedding adaptive neural networks into anomaly-based intrusion detection systems. The adaptiveness in the neural networks presents an approach that is dynamic towards identification of evolving cyber threats. The main challenges discussed are related to the inability of existing anomaly-based IDSs to handle both False Positives and ever-changing attack patterns. The research describes a novel system architecture that encompasses data preprocessing and feature selection. Specifically, the study leverages SelfOrganizing Incremental Neural Networks, Adaptive Resonance Theory, Growing Neural Gas, and Evolving Spiking Neural Networks, to create a flexible, adaptive detection model. This model trained and evaluated with a robust dataset regarding anomaly detection, considering advanced preprocessing, feature selection, and ensemble methods to reach an optimized model. The experimental results revealed very significant improvements in detection performance, such as 99.18% accuracy, 99.01% precision, 99.49% recall, and an F1-score of 99.23%. These results thereby indicate that adaptive neural networks may improve the effectiveness of IDS by rapid adaptation to new threats while improving detection accuracy and reducing false positives. In a nutshell, this research contributes to anomaly-based IDS by introducing adaptive neural networks as one way in which cybersecurity threats will always be changing, pointing at a way where network defenses can be more resistant and proactive.