Trends in Using Machine Learning for Network Anomaly Monitoring in Cloud Platforms

The article examines emerging trends in the application of machine learning methods for detecting network anomalies in cloud platforms, taking into account the influence of virtualization and the dynamics of infrastructure. The study is based on a systematization of publications addressing traffic and resource-metric monitoring, labeling scarcity, multilayer encapsulation, and the limited observability of distributed systems. The paper compares ensemble and hybrid architectures, convolutional and recurrent models, unsupervised autoencoders, as well as graph-based and contrastive approaches, enabling an assessment of how architectural choices affect anomaly-recognition robustness under distribution shifts and changes in packet structure. Special attention is given to the roles of temporal dependencies, node-interaction topology, and the effects of virtual-machine migration, which introduce distortions in input data and create opaque zones within virtual networks. The findings show that shifting from models relying on local features to spatiotemporal and graph-based architectures improves monitoring adaptability to cloud-infrastructure variability and partial labeling, although this transition is accompanied by increased model complexity and higher requirements for representation quality. The article may be of interest to professionals working in cloud-platform operations, network security, and reliability engineering for distributed systems.