Forecasting and Anomaly Detection on Application Metrics using LSTM
Antony Jerome, Tatsuya Ishii, Haichun Chen · 2018
This paper explores time series forecasting for application metrics, with the goal of performing anomaly detection and system resource management. To do so, we implemented a collection of models including an LSTM model, ETS and ARIMA, and compared their performances. Across different evaluation metrics, we observed that the LSTM model's forecasts performed well. We also constructed prediction bounds for the LSTM model's forecasts for identifying anomalies.