I need to use the SVM (used by the caret package [1] in R) in Java. One way would be to use rJava [2] and call R from Java but I will need to do this over 100,000 times so I can imagine that will take a while. Thus if I can access the SVM it uses without the overhead of R, everything will be much quicker!
Or an alternative would be to use an SVM which has a similar parameter training method to the one used in caret [3] as that seems to work relatively well. Other SVMs seem to favour one class on the prediction. The data used is class balanced, min-max normalised, though I haven't done too much investigation into parameter searching - RBF kernel, I tend to set the cost to between 1 and 100, and gamma as 1/num_features.
Cheers for any help :)
[link][2 comments]