Quite often in Data Science we deal with use cases where the penalty for False Negative can be huge. Examples can be missing to predict having a disease, denying loan based on false prediction, not acting on a cyber threat etc.
So the question is what kind of techniques exist to minimize or eliminate false negatives?
Tags:
https://network.youthmusic.org.uk/streamnow-richmond-tigers-vs-geel...
https://network.youthmusic.org.uk/liveespnrichmond-tigers-vs-geelon...
https://network.youthmusic.org.uk/live-tsn-afl-grand-final-2020-liv...
https://network.youthmusic.org.uk/livestreamfree-lrichmond-vs-geelo...
https://network.youthmusic.org.uk/watch-richmond-vs-geelong-live-st...
https://network.youthmusic.org.uk/total-sportek-richmond-vs-geelong...
https://network.youthmusic.org.uk/richmond-tigers-vs-geelong-cats-f...
How do those links relate to the topic?
This is not just about minimizing false negatives, but rather dealing with imbalanced classes. Some techniques to deal with this are:
- use the right evalualtion metrics - basically do not just use accuracy as it is rather deceiving for the imbalanced class problem; use AUC or f-score instead.
- down-sample or up-sample corresponding classes to have a more balanced set.
- there is a scale_pos_weight parameter in most of the decision tree based classifiers to adjust the weights of classes; play with this to get higher accuracy on your desired class.
Thanks Adil
Remember that a False Positive (FP) or False Negative (FN) is essentially caused by the output variable (the classification) overlapping in a low-dimension projection of the feature space of input variables. Thus, one way to reduce FP and FN is to introduce higher dimensions (more features, more tests, greater perspectives). Just like a doctor orders more tests (more test results = more data features to use in the diagnosis) when there is confusion in a diagnosis, a data scientist should "order more data" = more features, more dimensions. Look here: https://datamakespossible.westerndigital.com/five-ds-big-data-variety/
Thanks Kirk
© 2021 TechTarget, Inc.
Powered by
Badges | Report an Issue | Privacy Policy | Terms of Service
Most Popular Content on DSC
To not miss this type of content in the future, subscribe to our newsletter.
Other popular resources
Archives: 2008-2014 | 2015-2016 | 2017-2019 | Book 1 | Book 2 | More
Most popular articles