Time-frequency analysis and autoencoder approach for network traffic anomaly detection.

Area of Science:

• Cybersecurity and Network Analysis
• Machine Learning for Anomaly Detection

Background:

• Effective detection of network traffic anomalies is crucial for mitigating cyber threats.
• Existing methods may face challenges in scalability and real-time implementation for comprehensive cybersecurity.

Purpose of the Study:

• To develop and evaluate a hybrid approach integrating time-frequency analysis and autoencoders for robust network anomaly detection.
• To assess the scalability and real-time feasibility of the proposed model for practical cybersecurity applications.

Main Methods:

• Network traffic data (packet size, duration) underwent pre-processing and time-frequency analysis using Continuous Wavelet Transform (CWT), Discrete-Time Fourier Transform (DTFT), and Short-Time Fourier Transform (STFT).
• Extracted features were utilized to train an autoencoder model, with anomalies identified by deviations in reconstruction error.
• The hybrid model's performance was evaluated for scalability and real-time detection capabilities.

Main Results:

• The hybrid approach demonstrated good scalability for real-time cybersecurity implementations.
• The model achieved 95% detection accuracy, successfully identifying 72 network anomalies.
• Reconstruction error deviations effectively indicated anomalies such as spikes and irregular oscillations in network traffic.

Conclusions:

• The developed hybrid model is robust and scalable for real-time cybersecurity applications.
• The approach shows feasibility for deployment in practical cybersecurity scenarios.
• Further enhancements in autoencoder architectures could optimize performance in large-scale systems.