A Machine Learning-Based Anomaly Prediction Service for Software-Defined Networks.

Area of Science:

• Computer Science
• Network Security
• Machine Learning

Background:

• Software-defined networking (SDN) centralizes network control but lacks inherent anomaly detection capabilities.
• Network anomalies can indicate security threats, degrade performance, and compromise network integrity.
• Machine learning (ML) offers a potential solution for identifying complex traffic patterns and predicting threats.

Purpose of the Study:

• To propose and evaluate an ML-based service for predicting traffic anomalies in SDN environments.
• To enhance the security and performance of SDN by proactively identifying anomalous network traffic.
• To develop a robust system capable of detecting threats that traditional SDN controllers cannot.

Main Methods:

• Modeled a programmable data center with a signature-based intrusion detection system to create a comprehensive network traffic dataset.
• Pre-processed feature vectors for each flow request generated by forwarding elements.
• Trained a machine learning classifier using these feature vectors to predict traffic anomalies.
• Evaluated the proposed approach using the holdout cross-validation technique.

Main Results:

• The proposed ML-based service demonstrated high accuracy in predicting network traffic anomalies.
• Significant performance improvements were observed compared to baseline approaches (random prediction and zero rule).
• Average accuracy, precision, recall, and F-measure showed substantial gains, indicating the effectiveness of the ML approach.

Conclusions:

• The developed ML service is effective in predicting traffic anomalies within software-defined networks.
• This approach offers a substantial improvement over existing methods, enhancing network security and performance.
• The findings support the integration of ML for proactive threat detection in modern network infrastructures.