Re: About the feature weights

On 29/06/12 4:25 PM, myownstuff wrote:
>
> Hi, all
>
> I use LMT for classfication and use the relief attribute evaluation method
> combined with Ranker for feature selection
>
> I am confused about the meaning of the weight.
> For example, to classify {class0, class1}
>
> 0.38714285714286073 bkey
> 0.3557142857142888 vkey
> 0 victim
> -0.0357142857142857 afeature
> -0.036428571428571414 after
>
> Does it mean that "bkey" is the most important feature for class1, "after"
> is the most important feature for class0, and "victim" is not helpful for
> classification?
>
> Any response will be helpful.
> Thanks a lot!!
>
> WanChen
>
>

The output of ReliefF ranks the discriminatory power of the features 
from best to worst. Those with score <= 0 are definitely not useful for 
discriminating the classes according to the Relief method.