Orange Forum • View topic - Weights in Orange data

Weights in Orange data

General discussions about Orange and with Orange connected things (data mining, machine learning, bioinformatics...).

Weights in Orange data

Postby cs5080222 » Fri Aug 02, 2013 13:36

Hello
Can anyone explain in what instances can using the meta variable weight be beneficial? In the documentation It's mentioned that weight represents the importance of that data instance. This is not clear to me.

I am trying to classify a binary output(0/1).However the training data is highly uneven with 95% of the training instance being 0 and rest being 1. This leads to classification of most instances of test data as 0 by naive bayes learner. Can weight be used in this context by giving high weight to training instances with value 1 and vice versa?

Re: Weights in Orange data

Postby Ales » Fri Aug 09, 2013 16:36

cs5080222 wrote:I am trying to classify a binary output(0/1).However the training data is highly uneven with 95% of the training instance being 0 and rest being 1. This leads to classification of most instances of test data as 0 by naive bayes learner. Can weight be used in this context by giving high weight to training instances with value 1 and vice versa?

Yes, correcting for class imbalance is one of the intended uses of weights.


Return to General Discussions



cron