Error-Robust Distributed Denial of Service Attack Detection Based on an Average Common Feature Extraction Technique.

Area of Science:

• Cyber-Physical Systems Security
• Network Intrusion Detection
• Machine Learning Applications

Background:

• Cyber-Physical Systems (CPSs) face increasing advanced threats like Distributed Denial of Service (DDoS) attacks.
• Traditional machine learning-based Intrusion Detection Systems (IDSs) struggle with corrupted datasets, hindering effective DDoS attack detection.

Purpose of the Study:

• To propose a novel, error-robust multidimensional technique for enhanced DDoS attack detection in CPSs.
• To improve the resilience of IDSs against corrupted training data.

Main Methods:

• Utilized Higher Order Singular Value Decomposition (HOSVD) to filter average common features from datasets.
• Applied machine learning classifiers (Random Forest, Gradient Boosting) to classify filtered traffic data as legitimate or DDoS attacks.

Main Results:

• Achieved 98.94% accuracy, 97.70% detection rate, and 4.35% false alarm rate with 30% data corruption using Random Forest.
• Demonstrated superior performance over traditional methods in error-free conditions (99.87% accuracy, 99.86% detection rate, 0.16% false alarm rate with Gradient Boosting).

Conclusions:

• The proposed HOSVD-based technique offers a robust solution for DDoS attack detection in CPSs, outperforming existing methods.
• The approach significantly improves IDS performance, particularly in the presence of data corruption.