CODEWITHSUNDEEP
machine-learningscikit-learnartificial-intelligenceclassificationdata-science
Tobi M. Ottorony
Tobi M. Ottorony

Reputation: 129

Multiple Feature Sets of different Size for Classification

I have multiple different feature sets. These features are stored in dataframes. The Problem is, that they differ in size. My first featureset is a 512 long sized vector of float. The second featureset are 12 floats. And so on.

I would like to use different classifiers on all of this feature sets. I think that I cant put all the features into on single vector because the performance will break because the cure of dimensionality and the missing influence of the small vectors.

So I had an idea: Training multiple classificators on each featureset use these results as a new featureset for a final classificator.

Are there any standard ways of handling such problems?

Upvotes: 0

Views: 419

Answers (1)

stan0
stan0

Reputation: 11817

If I got it right that all your feature sets describe the same output then it makes most sense to actually predict based on all features. I would go with your first suggestion:

  • merging all features in a single vector, although long one, shouldn't be a problem for classifier like SVM - it's intended to work OK with many features.

Your other idea:

  • training a separate classifier for each feature set sound like adding additional complexity. Firs, you have one more abstraction to worry about. And second, the small/initial classifiers may not be that good if the have a limited amount of data small set of features (e.g. compared to a single SVM).

tl;dr: I would merge all features (if the different sets describe the same observations/variables, of course) and use them with an SVM.

Upvotes: 1

Related Questions