Adaptive Sampling Framework for Imbalanced DDoS Traffic Classification.

Area of Science:

• Network Security
• Machine Learning
• Data Science

Background:

• Class imbalance in network security data, especially for Distributed Denial of Service (DDoS) traffic, hinders accurate detection of underrepresented attack types.
• Existing machine learning and deep learning models struggle with imbalanced datasets, leading to performance degradation and ineffective defense strategies.

Purpose of the Study:

• To propose and evaluate an adaptive sampling strategy to address class imbalance in DDoS traffic classification.
• To improve the detection of minority classes in imbalanced network security datasets.

Main Methods:

• Developed an adaptive sampling strategy combining oversampling and undersampling techniques to rebalance the dataset at the data level.
• Evaluated the proposed method on benchmark DDoS traffic datasets.

Main Results:

• The adaptive sampling strategy demonstrated improved classification performance compared to baseline models and conventional sampling methods.
• Key metrics such as accuracy, recall, and F1-score were enhanced, particularly for minority class detection.
• The approach improved the reliability of sensor-driven security systems by enhancing anomaly detection capabilities.

Conclusions:

• The proposed adaptive sampling method offers a robust and adaptable solution for imbalanced data classification in network security.
• This technique is particularly effective for improving minority class detection in DDoS traffic classification.
• The findings have potential applications in simulated sensor environments requiring essential anomaly detection.