BALANCE: Bayesian Linear Attribution for Root Cause Localization

BALANCE applies explainability techniques to a practical problem: The paper demonstrates how explainable AI (XAI) methods—typically used to understand AI model decisions—can be repurposed for root cause analysis in computer systems, showing that XAI frameworks have broader applications beyond model transparency.

The method handles interconnected causes effectively: BALANCE's Bayesian feature selection component specifically accounts for correlations between potential root causes, addressing a real challenge in system diagnostics where multiple factors often influence each other rather than acting independently.

Strong real-world performance across diverse scenarios: The method was tested on three different types of system failures (SQL queries, container faults, and fault diagnosis), outperforming existing approaches, suggesting the approach generalizes well to different types of technical problems rather than being limited to one domain.

Transparency in system diagnostics supports responsible operations: By providing attribution scores that explain which components caused system anomalies, BALANCE enables operators to understand why a system failed, not just that it failed—supporting more informed and accountable decision-making in critical system maintenance.

Addresses a gap between detection and recovery: The paper identifies that while systems are good at detecting faults, pinpointing root causes remains challenging; BALANCE bridges this gap, potentially reducing downtime and preventing repeated failures through better understanding of underlying issues.