A TEMPORAL LOGICAL ATTENTION NETWORK APPROACH TO ANOMALY DETECTION IN DISTRIBUTED SYSTEMS LOGS

This paper presents the Temporal Logical Attention Network, a novel approach to anomaly detection in distributed systems logs. As distributed systems become increasingly integral to modern applications, the complexity and volume of log data generated pose significant challenges for effective monitoring and analysis. Traditional methods for anomaly detection, such as rule-based and statistical techniques, often fall short in addressing the dynamic nature of log data, resulting in high false positive rates and inadequate detection of subtle anomalies. TLAN leverages deep learning, specifically attention mechanisms, to capture temporal dependencies and logical relationships within log data. By embedding log entries into a dense vector space and applying temporal encoding, TLAN identifies significant patterns over time, enhancing the accuracy of anomaly detection. The model focuses on relevant log entries, allowing it to prioritize critical information while minimizing the influence of less significant data. Through rigorous experimentation on multiple datasets, TLAN demonstrated superior performance compared to traditional and state-of-the-art models, achieving high precision, recall, and F1-scores. The findings underscore TLAN's effectiveness in identifying anomalies that may indicate underlying issues, such as security breaches or system failures. This research contributes to the evolving landscape of anomaly detection techniques, highlighting the importance of integrating advanced machine learning approaches in managing distributed systems logs. Ultimately, TLAN represents a significant advancement in the field, offering organizations robust tools for enhancing the security and reliability of their distributed environments.