Randomized trial evaluates a new clustering method for KPIs, suggesting improved threat detection and alert management.
Key Points
This research aims to develop a scalable method for clustering large-scale KPIs to enhance cybersecurity operations. The focus is on balancing clustering quality and computational complexity.
Proposed the SubCluster method for clustering KPIs effectively.
Utilized coarse-grained clustering with a nearest neighbor-based graph for initial grouping.
Extracted representative subsequences for fine-grained clustering.
Demonstrated effective clustering performance with reduced computational overhead.
Improved detection of system failures by 30% based on experimental data.
Reduced redundant alarms in real-world datasets significantly.