LUCID – A Lightweight Deep Learning Approach for DDoS Detection

Abstract

Distributed Denial of Service (DDoS) attacks remain a persistent threat to the availability and reliability of online services, frequently overwhelming target networks with large volumes of malicious traffic. Traditional detection mechanisms often suffer from high resource consumption, limited adaptability to evolving attack vectors, and poor scalability in real-time environments. To address these limitations, this project proposes Lightweight Unsupervised Clustering for Intrusion Detection (LUCID) a novel, lightweight, and scalable DDoS detection system that leverages the feature extraction capabilities of Convolutional Neural Networks (CNNs) in an unsupervised learning framework. LUCID introduces a unique data preprocessing pipeline that transforms raw network traffic data into structured image like representations suitable for CNN input, effectively capturing temporal and spatial patterns in traffic flows. By eliminating the need for labelled training data, LUCID enables detection of both known and previously unseen (zero-day) attack patterns. The system is designed for real-time deployment, offering low computational overhead and minimal latency, making it suitable for edge devices and high-throughput environments. Overall, LUCID represents a significant advancement in intelligent network defines, combining efficiency, robustness, and interpretability in a single framework tailored for modern cyber security challenges.