A Comparative Analysis of Statistical Anomaly Detection Methods for Cloud Service Monitoring: A Simulation-Based Evaluation Framework

Cloud service reliability depends critically on effective anomaly detection in system metrics. This paper presents a comprehensive simulation-based evaluation framework for comparing statistical anomaly detection algorithms in cloud environments. We implement and evaluate four statistical detection methods (Z-Score, Modified Z-Score, EWMA, and Threshold-based) across four key cloud metrics (CPU usage, memory usage, network I/O, and response time) using five distinct anomaly patterns (spike, dip, level shift, trend change, and collective anomalies). Our experimental results reveal that Threshold-based detection achieves the highest overall F1-score (0.142), while Modified Z-Score detection demonstrates superior precision (0.209). The study provides empirical insights into algorithm performance trade-offs and introduces a reusable simulation framework for systematic evaluation of anomaly detection methods in cloud computing contexts.