Anomaly Transformer-based System Log Anomaly Detection

Modern systems, such as cloud-centric applications with service-oriented architectures, generate massive system log files that capture essential runtime information and events. These logs are crucial for monitoring system performance and identifying potential issues. Traditional methods, such as rule-based techniques and Long Short-Term Memory (LSTM) networks, have limitations in handling the complexity and volume of system log data. This paper presents a novel Anomaly Transformer-based system log detection architecture featuring a log parser, feature extractor, sequence generator, and transformer-based anomaly detection model. Our approach achieves superior performance in detecting anomalies in the system log file, as demonstrated by outperforming existing system log-based anomaly detection techniques on the HDFS dataset, offering a more accurate solution for system and software monitoring.