Tools
Tools
Allahabadi, Jai (2024) Hybrid Anomaly Detection Framework for Kubernetes Environment. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (1MB) | Preview |
Abstract
Over 60% of the enterprises have adopted Kubernetes and as per CNCF survey, the adoption rates have been increased to 96%. With such a high adoption rate, security concerns also arise exponentially. The market size for K8s security will be projected to reach $27.19 billion by 2032. Hence, the need to delve into the security of the K8s has become the need of the hour. With the advancement of artificial intelligence, the intrusion of the AI algorithms for anomaly detection has been significantly increasing. This paper builds upon the hybrid model that employs Long-Short Term Memory (LSTM), custom attention layer and Transformer network, for detection of anomalies along with the help of feature engineering techniques i.e., Principal Component Analysis (PCA) and Autoencoders. The hybrid model has been trained using traditional and Model-Agnostic Meta Learning (MAML) methods. NSL KDD and Kubernetes based attacks datasets have been employed in this research. Extensive experiments have been stemmed from an intent to explore the synergy between feature engineering techniques and training methods, with the conclusion that hybrid model trained on Autoencoder features data using traditional method surpasses with 98% accuracy and 0.98 F1 score. However, training the hybrid model trained using MAML reduces the training time up to 99% compared to traditional method.
Actions (login required)
 |
View Item |