Design and Implementation of a Semi-supervised Anomaly Log Detection Model DDA
Yu‐Min Wang, Zixiang Ji · 2021
In this paper, a semi-supervised log anomaly detection model DDA was proposed. By modeling and deeply optimizing the log extraction function of Drain and the log sequence anomaly detection function of DeepLog and by combining the first-order outlier detection algorithm of parameters, an end-to-end log anomaly detection model - DDA was constructed. The DDA model was divided into front-end and back-end, where the front-end is responsible for preprocessing the logs such as specification, extraction of parameters and templates, and clustering, and the back-end implements the core functions of log anomaly detection. Experiments demonstrated the superior performance of DDA in terms of anomaly detection rate.