MULTIVARIATE TIME-SERIES ANOMALY DETECTION IN DATA CENTERS: A HYBRID APPROACH USING AUTOENCODERS AND PRINCIPAL COMPONENT ANALYSIS

This article presents a novel approach to detecting anomalies in Data Center infrastructure by combining autoencoder neural networks with Principal Component Analysis (PCA) for multivariate time-series data.The proposed framework addresses the challenges of detecting complex temporal patterns across multiple infrastructure metrics while maintaining computational efficiency.The method leverages deep autoencoders to learn normal behavioral patterns from historical Data Center metrics, including temperature, power consumption, CPU usage, and network traffic, while utilizing PCA for effective dimensionality reduction.The article demonstrates superior performance compared to traditional statistical methods and standalone deep learning approaches in both accuracy and computational efficiency.Experimental results on real-world Data Center infrastructure data show that our hybrid approach successfully identifies anomalies across various operational scenarios while minimizing false positives.The article offers practical implications for improving Data Center reliability and operational efficiency through early detection of potential infrastructure issues.Furthermore, the framework's scalability and adaptability make it suitable for deployment in diverse Data Center environments.