A NOVEL DEEP BELIEF-BASED MODEL FOR THE INTRUSION DETECTION OF NETWORK TRAFFIC

ABSTRACT:

The fast expansion of networked systems, combined with the pervasive reliance on the internet, has raised worries about network security, needing new defense methods. Intrusion Detection Systems (IDS) use a variety of ways to distinguish between legitimate and malicious network traffic, including rule-based, signature-based, anomaly detection, and machine learning approaches. While signature-based IDS excel at detecting known threats, they struggle with novel attacks, prompting the development of anomaly-based IDS and machine learning methods such as Random Forest and Logistic Regression, among others. However, these techniques have scalability and computational complexity challenges. Cybersecurity remains a significant concern for organizations due to continual cyber-attack threats, which drives ongoing development into intrusion detection systems. Deep Learning (DL)-based IDSs have gained popularity due to their deep feature learning capabilities, while being resource-intensive. To overcome computational problems, this research provides an optimized deep belief-based model that combines the Genetic Algorithm, Particle Swarm Optimization, and Probabilistic Neural Network (GePP-Dbnet). This model seeks to find a balance between accuracy, training duration, and false alarm rates while identifying a diverse set of threat classes. Validation will be carried out using the benchmark datasets NSL-KDD and CSE-CIC-IDS2018, which provide realistic scenarios for assessing the model's effectiveness.

Keywords: Intrusion Detection, Attacks, Deep Learning, Deep Belief, Dataset.