Anomaly Detection in Cloud Computing Workloads Based on Resource Usage


Jahani A., Amirkhizi P. J.

JOURNAL OF NETWORK AND SYSTEMS MANAGEMENT, cilt.33, sa.4, 2025 (SCI-Expanded, Scopus)

Özet

Cloud computing services face increasing security threats, which is a challenging problem. The existing anomaly detection methods struggle with multi-metric correlations and missing data. To address these challenges, this paper proposes Pattern-AD, a novel anomaly detection method that models attacks as anomalies against the system's normal states. Unlike traditional approaches, Pattern-AD extracts frequent patterns using the Apriori data mining algorithm, offering flexibility regardless of pattern length. Evaluated on the GWA-T-12 dataset (1750 VMs), the proposed Pattern-AD achieves 99.98% accuracy, outperforming KNN and Isolation Forest by >=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge$$\end{document} 49%, with an event processing latency of 1.5 s. Most importantly, it maintains 96.55% accuracy even with 20% missing data, a capability unmatched by deep learning alternatives. This provides cloud operators with an interpretable and lightweight solution for anomaly detection.