machine learning - One Class Classifier Training -
i'm working on classification problem have data 1 class, wanna classify between "target"class against other possibilities "outlier" class. therefore, intend use 1 class classifier or libsvm classifier. question here is:
do need provide training data "outlier" class? if yes, there way around have data target class.
one way achieve samples positive class density estimation. can either fit parametric model data (for example, multi-variate normal) or use kernel density estimator (a little bit one-class version of nearest neighbour, kernelised distance metric). then, evaluate probability of new data under learned model, , if it's sufficiently low, reject member of class.
Comments
Post a Comment