Cost sensitive learning

Submitted by buki on Tue, 01/24/2012 - 14:00

General Neural Networks

Hi,

if I want to apply cost sensitive learning for a NN I simply implement my on ErrorFunction which gives each misclassification (false negative, false positive) a different weighting, right? E.g. for misclassifying a positive example I give a quadratic erorr and for a negative example I stick with atan.

Or is my approach too simplistic?

Regards,

Andi

Submitted by gtakacs on Wed, 01/25/2012 - 08:13.

Hi,

I found that BasicMLDataPairs have a Significance property. By default this value is set to 1. I guess this should be the weight of a learning example. I may be wrong, but as i saw, pretty much every learning algorithm ignores this property.

Regards,
Gabor