Abstract
We built a new framework to evaluate a production failure predictive model performance using a
set of healthy non failure data. This framework gives us the capability to quickly access the model
performance as soon as we deploy it into production. In the past, in order to access a failure
predictive model performance, ones need to have some number of failures to happened in the field
and then looking back in time if prediction was ever made for those failures. If the true positive
windows is 60 days then you have to wait more than 60 days to be able to access a field model
performance. We can shorten that waiting period by leveraging healthy data that we collected from
the field.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.
Recommended Citation
INC, HP, "A FRAMEWORK TO MONITOR A FAILURE PREDICTIVE MACHINE LEARNING MODELS PERFORMANCE WITH NON-FAILURE DATA", Technical Disclosure Commons, (March 23, 2020)
https://www.tdcommons.org/dpubs_series/3039