End-to-End Threat Hunting with a Novel Multiclass Dataset for Intelligent Intrusion Detection

With the rapid evolution of technology, interconnected networks have become essential across all digital systems, enabling seamless global integration. While these advancements enhance system capabilities and user experiences, they also introduce significant cybersecurity risks. We propose a comprehensive threathunting framework spanning data generation to intelligent intrusion detection to address this. The objective is to design a robust intrusion detection system (IDS) capable of distinguishing normal from malicious traffic and classifying threats into 15 distinct cyberattack categories, including MITM ARP Spoofing, SSH Brute Force, SQL Injection, and DDoS variants. Our approach introduces a novel, largescale multiclass dataset for modern network environments. We evaluated this dataset using seven supervised learning models: Naïve Bayes (NB), Logistic Regression (LR), Random Forest (RF), Decision Trees (DT), Feedforward Neural Networks (FNN), Multi-Layer Perceptron (MLP), Deep Convolutional Neural Networks (DCNN), and Shallow Convolutional Neural Networks (SCNN). Among these, DT-based models demonstrated the highest performance, achieving 99.9% accuracy in detection and classification, with a minimal prediction latency of 1.1 microseconds. This research contributes a scalable and effective IDS framework to enhance the security posture of communication networks against diverse and evolving cyber threats.