Search code examples
algorithmmachine-learningclassificationlibsvm

Best one class classifier for small sample size

I have a high-(600-)dimensional dataset and I have a very small sample size (+- 100 datapoints) from one class. The nature of data make it very hard to generate outliers, so I need to I need to train a one class classifier on these data.

Do you have experience with one class classification? What is the best algorithm for small sample size and high dimensionality? I was thinking about using one class SVM from libsvm library, is it a good choice?

Thanks for advice

Solution

  • If you wanted to take the path of neural networks then:

    For high dimensionality single class classification you probably would want a needle-in-haystack style training regiment.

    I believe the most notable of these would be the adaptive gradient.

    Otherwise, high dimensionality is handled efficiently by SVMs and the fact that outliers are unlikely further strengthens the case for using a boundary-based algorithm.