A Deep Learning-Based Abnormal Detection Study Using Cloud Time Series

With the increasing complexity of cloud-native environments, timely detection of anomalies in system performance has become crucial. Cloud infrastructures produce large volumes of multivariate time-series data across various metrics including CPU, memory, and network I/O. Traditional statistical methods struggle with the nonlinear dependencies in such data, leading to a shift toward unsupervised machine learning techniques. This study investigates the effectiveness of two deep learning-based anomaly detection models—LSTM Autoencoder and GMM-GRU-VAE (GRU-based Gaussian Mixture Variational Autoencoder)—using real-world cloud infrastructure data. This study conducts an in-depth analysis of the architectures of the two models and compares their strengths, limitations, and performance in light of the specific requirements of cloud environments. By doing so, it aims to provide both theoretical and practical foundations for selecting the most suitable model for various cloud anomaly detection scenarios.