Thomas Lumley • 8/29/2015

Net Reclassification Index: surprisingly weird.

This in-depth article analyzes the Net Reclassification Index (NRI), a metric used to assess improvement in predictive models when new information is added. It explains the NRI's formula, discusses the intuitive expectation that bias disappears with an independent test dataset, and reveals why this isn't the case. The author uses a detailed example to demonstrate how the NRI can produce misleading positive results even when a new prediction rule is worse than the old one.

0 comments

#classification #statistics #prediction models