Self-learning Machine Method for Anomaly Detection in Real Time Data

Abstract

Cloud service monitoring requires robustness. Usually the service and its Key Performance Indicators (KPIs) are growing incongruently along with the growth of cloud infrastructure, dependencies and feature set. Even with validated software, physical misconfiguration can cause the service failure and may lead to service outage. That is why it is important to automatically detect any abnormal behavior and integrate it with the Event Management System (EMS) for proper and timely escalation. This paper presents a lightweight anomaly detection method, which is able to identify the pattern of metric's behavior and will be able to adjust itself to possible pattern modification caused by either new service releases and/or natural changes of utilization.