I have a set of relatively long ($\sim 1000$) binary features with scalar values $[0-10]$ attached to them. My aim is to write a predictor that learns to map the features to the $[0-10]$ interval to predict new features when given a new binary vector. I used SVM and Lasso with leave-one-out performance analysis, but both always end up predicting the mean value of the distribution (correlates to the histogram of all the feature - scalar distribution). The histograms are also rather norm / Rayleigh distributions. Suggestions for algorithms / feature space mapping? My main problem is that I am dealing with binary features for the first time.

  • $\begingroup$ Well, there is always the possibility that your features do not give you useful informations about your scalar values... Have you tried to do some exploratory analysis of the features? For example, looking if there is correlation between them, comparing the mean value for different subsets, things like that. That can give you a hint about how informative your features are. $\endgroup$
    – Jundiaius
    May 12 '14 at 11:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.