Abstract:
|
Nowadays large companies have many systems and applications built as web-based services, to ensure undisrupted operations, one needs to closely monitor various metrics, such as total number of users, response time, or usages. Detecting anomalies in Key metrics and making timely troubleshooting is crucial to prevent potential failure on relevant applications. This paper proposed automated ensemble anomaly detection methods composed by creating more than 15 different detectors, and over 5 different machine learning detection models, which is designed for large-scale metrics to be detected and lack of anomaly labeling. We compared our methods with published research work done by well-known professor Dan Pei and his students, our results are in competitive positions, and have been proved practical working, accurate, and efficient in real-world cases.
|