n-Pipeline Log Anomaly Detection Drift Mitigation

A significant obstacle to the industrialization of log anomaly detection method usage for production system monitoring is the difficulty managing dynamic system state. Production systems continuously evolve, necessitating adaptation by the models that build representations of their behavior. Previous work to mitigate the impact of log drift has focused primarily on feedback mechanisms. These methods, however, are manual and have numerous complexities. This study introduces a new mitigation approach in the form of n-pipelines. We compare the performance of n-pipelines (alternating between active and passive states) to traditional log anomaly detection workflows, assessing their performance and highlighting the benefits of their use.