I am investigating time-series anomaly detection for industrial machine data, and have stumbled upon GE Predix. It seems like a promising tool, however, I am not familiar with their approaches for anomaly detection.

Does anyone have experience with GE Predix, in the anomaly detection space? How would one rate its anomaly detection capability? What technical approaches (algorithms) does it use? Any other literature about GE Predix that is openly available that I have missed (besides References below)? Are there any available benchmarks (e.g., versus Splunk + Prelert)? Who are its competitors?

1) https://www.predix.com/sites/default/files/predixplatformbrief-march_2016.pdf
2) https://www.predix.com/sites/default/files/GE-Software-Modernizing-Machine-to-Machine-Interactions.pdf


1 Answer 1


(full disclosure, I do work for GE)

GE itself has been using their GE Predix Anomaly detection algorithms to establish a relationship between anomalies & machine failure for their factories, in an effort to establish a preventive maintenance framework. The algorithms have been heavily tested to catch any point anomalies from a machine data. Go ahead and test it out for your datasets: The algorithms are available for registered developers (in catalog) to use in their applications.

Technical approaches (algorithms): Univariate/Multi-variate anomaly detection algorithms. It detect Shift anomalies , Unexpected trends in physical system behaviors, Anomalous patterns & detect changes in relationship among different variables. It also detects data with very high dimensions (>100). Mann-Kendall, Mann-Whitney, Kernel density estimate etc are some of the key algorithms used.

  • $\begingroup$ Thank you, @Lothar, this is helpful. The references I have above seem to be at a very high level with minimal technical material; Do you know of any published comparisons of GE Predix's capabilities (algorithm performance, comparisons, etc) in literature? Something like Numenta's NAB results would be great. $\endgroup$
    – ximiki
    Jul 8, 2016 at 15:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.