Dynamic Real Time Framework for Abnormal Detection of IMS Core in Kubernetes Cloud

Abstract

In the ever-evolving telecommunications sector, advancing from 5G towards 6G, maintaining the security of core infrastructures has become supreme. This study addresses the critical need for proactive and real-time anomaly detection within cloud-native environments. Leveraging cloud-native implementations within Kubernetes clusters, our framework utilizes advanced machine learning techniques to analyze data from applications and clusters. Specifically, this paper introduces a novel integration of k-means clustering and Long Short-Term Memory (LSTM) models for real-time anomaly detection in Kubernetes-based cloud-native environments, offering a unified framework capable of addressing both global and local anomalies across multiple layers of the IP Multimedia Subsystem (IMS) core. Existing research on anomaly detection in Kubernetes environments often focuses on specific application layers or isolated metrics, lacking a comprehensive solution that addresses the multidimensional and dynamic nature of IMS core anomalies across application, pod, and node levels in real-time. By employing k-means clustering and LSTM models, our approach achieves approximately 90% accuracy in anomaly detection. Extensive experiments with various model versions demonstrate the effectiveness of the framework, ensuring robust security and reliability for next-generation telecom networks.