DeepSipi: A Log Anomaly Detection Method with Events and Variables
Jinhui Yuan, Hongwei Zhou, Lianghui Li, Guang Chen, Fuling Li · 2023
Existing log anomaly detection algorithms often take log events as input. However, variables in log entries are also valuable for anomaly detection. To this end, we propose a log anomaly detection method that combines variables and events, which we call DeepSipi. During log preprocessing, DeepSipi constructs log event sequences and corresponding log sequences, and uses LSTM to train and detect them. We have constructed a prototype system based on the Keras neural network framework. Experiments have shown that DeepSiPi implemented anomaly detection from two dimensions of variables and events, and improved the recall rate and F1 values.