Robust and Scalable Deep Learning Framework for Anomaly Detection in Large-Scale Network Security Systems
Gurbakhsis Singh · International Journal of Intelligent Systems and Applications in Engineering · 2024
With the rising complexity of cyber threats, scalable and intelligent intrusion detection systems are critical for safeguarding large-scale networks. Traditional signature-based methods often miss zero-day attacks, while classic machine learning struggles with high-dimensional traffic data. This study presents a deep learning framework for accurate anomaly detection using the CICIDS2018 dataset, which includes diverse modern attack patterns. The proposed system employs Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and a hybrid CNN-LSTM model to extract both spatial and temporal features from traffic data. Among the models, CNN-LSTM achieved the highest accuracy of 98.8%, surpassing CNN (97.9%) and LSTM (98.2%). Classical models like Support Vector Machine (SVM) and K-Nearest Neighbours (KNN) lagged behind, each scoring 91.8%. These findings highlight the superiority of deep learning in detecting complex intrusions. Future work will focus on real-time implementation, reduced computational costs, and the adoption of explainable AI for better transparency and usability in IoT and edge computing scenarios. DOI: https://doi.org/10.17762/ijisae.v12i17s.7685