I aim to have a large portion of negative instances for my SVM models. However I cannot get any random set to use it as negative since in those sets some instances include those colors as well. How can I gather such a large negative set preventing that inclusion problem as much as possible? I also do not want to use other classes as negative since collecting random negatives empirically works better.
Maybe try a one-class SVM instead? Or try to generate outliers / negative examples around your positive class: http://machinelearning.wustl.edu/mlpapers/paper_files/TaxD01.pdf
I already tried one-class svm but did not work since the data is so diverse. Thanks :)
You may do your classification job in another feature space, where data have better clustering performance.
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification