I have been looking for research articles that target the question "why did an anomaly occur?", however, most of the papers that I found are related to "if an anomaly occurred?". I searched Google Scholar with key terms (both independently and together as phrases) like, "root-cause analysis of anomalies", "anomaly diagnosis", "causal analysis of anomalies", "causal inference for anomalies", all in the context of time-series and multivariate time-series, but couldn't find much.
I am really interested in knowing state-of-the-art, specifically in the domain of anomaly diagnosis (or root-cause analysis) in the context of multivariate time series. Can experts in the domain refer me to some good research articles in the area? For me, anomalies are spikes and volatility in the time-series.
You might read my paper on HFACS posted on this site for starters. I have been extensively involved in RCA for decades. There really are no "new" methods, just a lot of commercial variations on the few key factors. I think that the HFACS model does the best for analysis, but some other techniques may be better for investigations.
Thanks Steven for referring me to HFACS, I wasn't aware of it. It is interesting that while there is so much research on the detection part, people are hardly working on developing data-driven ways for the reasoning of those anomalies.
Milan Jain it is easier to focus on (and blame) THINGS than the people and their actions that allow those things to cause harm. The result of my evidential report to a senior management team received immediate praise and subsequent hiding, as it pointed out the MOST of the causes (and solutions) to serious problems were the responsibility of those top managers!
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series