Our paper, “Conditional Attribution for Root Cause Analysis in Time-Series Anomaly Detection,” by Shashank Mishra, Karan Patil, Cedric Schockaert, Didier Stricker, and Jason Rambach, was accepted for an oral presentation at ECML PKDD 2026, held in Naples, Italy, from September 7 to 11, 2026. Shashank Mishra presented the work at the conference.
The paper introduces a novel conditional attribution framework for identifying the sensors responsible for anomalies in multivariate time-series data. By comparing anomalous observations with contextually similar normal system states, the proposed approach preserves dependencies between sensors and provides more reliable explanations. Experiments on industrial benchmark datasets demonstrate improvements in root cause identification, temporal localization, and robustness.
Paper: https://link.springer.com/chapter/10.1007/978-3-032-37685-5_10
Contact: Shashank Mishra