An Enhanced Intrusion Detection System using Hybrid CNN-LSTM Architecture

Abstract

The rapid transition toward interconnected digital ecosystems and the rise of the Internet of Things (IoT) have necessitated the development of advanced Intrusion Detection Systems (IDS) capable of identifying complex, non-linear attack patterns in real-time. Traditional machine learning models and signature-based systems, while foundational, often struggle with the high-velocity nature of modern network traffic and are inherently incapable of identifying zero-day exploits. This research proposes a hybrid deep learning architecture that integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks to capture both spatial and temporal characteristics of network flows. In this framework, the CNN component extracts hierarchical spatial features and localized protocol patterns, while the LSTM component models long-term temporal dependencies inherent in sequential packet sequences. Experimental evaluation using benchmark datasets demonstrates that the hybrid CNN-LSTM model significantly outperforms standalone architectures, achieving validation accuracies exceeding 98.5% and a multiclass F1-score above 96%. This research underscores the importance of spatiotemporal fusion in developing resilient, future-ready cybersecurity infrastructures.

Keywords