Anomaly Detection for Key Performance Indicators Through Machine Learning
Jia Shun Shi, Gang He, Xinwen Liu · 2018
The data center and servers generates a large amount of monitoring data and logs every day. With the rapid development of the Internet, web services have penetrated into all areas of society. The stability of the Web services depends on IT operations to guarantee, operations staff through the monitoring of various key performance indicators (KPIs) to judge whether the Web service is stable or not. We want to use the statistical and machine learning methods to detect anomalous. In this paper, we split the problem into two parts. In the first place, we use time series analysis method such as Triple Order Exponential Smoothing (Holt-Winters) and ARIMA model and regression-based machine learning techniques such as Gradient Boosting Regression Trees (GBRT) and Long Short-Term Memory (LSTM) to predict the value at the next point in the time series. After that, we set the anomaly detection rule. Finally we compare the predicted value with actual value to determine whether the current point is anomalous or not.