A Convolutional Neural Network for Improved Anomaly-Based Network Intrusion Detection.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• The increasing reliance on technology necessitates robust cybersecurity measures.
• Network intrusion detection systems (NIDS) are critical for securing computer networks.
• Anomaly detection is a key approach for identifying sophisticated cyber attacks.

Purpose of the Study:

• To propose two novel deep learning models for network attack classification.
• To address both binary and multiclass classification of network intrusions.
• To improve the effectiveness of anomaly detection-based NIDS.

Main Methods:

• Utilized a convolutional neural network (CNN) architecture for attack classification.
• Developed a hybrid two-step preprocessing approach for feature generation.
• Combined dimensionality reduction and feature engineering with deep feature synthesis.

Main Results:

• The proposed models demonstrated strong performance on benchmark datasets (NSL-KDD and UNSW-NB15).
• Achieved high accuracy and recall in classifying network attacks.
• Outperformed existing deep learning and state-of-the-art classification models in literature.

Conclusions:

• The developed deep learning models offer an effective solution for network intrusion detection.
• The hybrid feature engineering approach enhances model performance.
• These models contribute to advancing cybersecurity frameworks through improved network attack classification.