DeepContainer: A Deep Learning-based Framework for Real-time Anomaly Detection in Cloud-Native Container Environments
Ke Xiong, Zhonghao Wu, Xuzhong Jia
Abstract
This paper presents DeepContainer, a novel deep learning-based framework for real-time anomaly detection in cloud-native container environments. The proposed framework addresses critical security challenges in containerized infrastructures through an innovative integration of neural network architectures and automated response mechanisms. DeepContainer implements a multi-layered detection approach, combining feature engineering techniques with optimized deep learning models to identify security anomalies across diverse container workloads. The system architecture incorporates specialized components for real-time data collection, processing, and analysis, achieving a detection accuracy of 96.8% with an average response latency of 7.3ms. Experimental evaluation in large-scale Kubernetes environments demonstrates significant performance improvements over existing solutions, including a 39.7% reduction in detection latency and a 25.5% decrease in resource utilization. The framework maintains linear scalability up to 10,000 monitored containers while achieving a false positive rate of 0.008. Comprehensive security testing validates the system's effectiveness across multiple attack vectors, including network-based attacks, resource exhaustion attempts, and access violations. Through automated response capabilities and sophisticated threat classification mechanisms, DeepContainer establishes a robust security foundation for modern containerized applications, addressing critical gaps in existing container security solutions.