below 50% binary classification accuracy

✉️ Mailing List Archive (read-only)
1

External Email - Use Caution

Hi everyone,
I'm doing classification (only two EEG data classes with equal sizes). I'm
using Random Forest with CSP and Logistic Regression with CSP. With CSP and
Logistic Regression I got about 65% classification results, whereas with
Random Forest + CSP I get less than 50% classification accuracy. I think
this is wrong, as it should be higher or equal to 50%. So what is the
possible reason for this? Can I simply say that in case I have 45% accuracy
then this can be calculated as 100-45=55% accuracy? Can I say that this
Random Forest algorithm is simply bad algorithm for my data?

Thanks all for any suggestions
I'm ready to share the data and the code if any one can help
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190725/df6a11b0/attachment.html

2

External Email - Use Caution

Hi,

I cannot say much about the data but you certainly cannot do 100-45=55%
accuracy :slight_smile:

It's possible the random forest is overfitting because of non-linearities.
I'd venture to guess you have a small sample size.

Mainak