DOC-IDS: A Deep Learning-Based Method for Feature Extraction and Anomaly Detection in Network Traffic.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• The increasing sophistication of cyberattacks necessitates advanced anomaly-based intrusion detection systems.
• Current machine learning and deep learning methods often require extensive feature engineering and labeled data, posing significant challenges.

Purpose of the Study:

• To propose a novel deep learning model, DOC-IDS, for automated feature extraction and anomaly detection in network traffic.
• To address the limitations of existing methods by reducing the need for manual feature design and labeled data.

Main Methods:

• DOC-IDS utilizes a deep one-class classification approach, integrating one-dimensional convolutional neural networks and an autoencoder.
• The model employs three distinct loss functions during training and leverages multi-class labeled traffic for enhanced feature extraction.
• Variance minimization in the feature space is applied to improve the discrimination between normal and abnormal traffic.

Main Results:

• DOC-IDS demonstrates higher anomaly detection performance compared to traditional methods.
• The model successfully automates feature extraction and anomaly detection, significantly reducing the manual workload.
• Experiments confirm the enhanced ability of DOC-IDS to discriminate between normal and abnormal network traffic.

Conclusions:

• DOC-IDS presents an effective and efficient solution for anomaly-based intrusion detection.
• The proposed model simplifies the process of implementing intrusion detection systems by automating key feature engineering steps.
• This research contributes to advancing the field of network security through innovative deep learning applications.