SMV question - all the same predictions

danjeharry

Hi all,

I'm fairly new to data mining so am trying to understand some of the results I'm getting. I've been using the libsvm operator available to predict my data. I used an epsilon regression with a radial basis kernel (b/c there are non-linear relationships) and optimized the parameters by randomly sampling 5000 examples out of a total 25000 or so. I then went back and trained the SVM on the whole training set(22000) and tested on the OOS set (3000) or so. I think I may have done something wrong because I get very strange results, e.g. the predictions the SVM model gives me for my out of sample data is all the same (e.g. -.5 for all my OOS data). This is obviously incorrect, but I am unsure how to interpret the results to track down my error. Any ideas would be great.

Thanks,
danjeharry