FastTransLog: A Log-based Anomaly Detection Method based on Fastformer

Abstract

In daily operation, the log, as one of the most important information to record the status of the system, is a part of the content we need to pay attention to. Therefore, there are a lot of research on log anomaly detection. However, through our learning, it was found that these models based on log parsing obviously had the following shortcomings: 1) Easily affected by out-of-vocabulary words, accuracy of the result is decreased. And 2) taking a long time to calculate. In order to remedy the above defects, I propose a log exception detection method based on Fast-Former namely FastTransLog in this paper, and abandon the traditional log parsing process. Through this method, not only the algorithm speed is greatly improved, but also the accuracy of the datasets is up to 99%.