Feature vectors are generally distributed in a way that their
intrinsic similarities cannot be easily quantified in the input
space as they are non-linearly separable [31]. For this reason,we use a non-linear transformation () to map each feature
xti
2 Ft to a high-dimensional non-linear feature space where
the feature vectors and Support Vectors are separable. The
mapping (xti
) need not to be computed explicitly as we
can implicitly quantify the feature similarity in the feature
space (via the so-called kernel trick). Such a feature similarity
allows us to determine the Support Vectors by generating a
hypersphere with center ct and radius Rt that encloses feature
vectors of the same clusters in the feature space. The Support
Vectors are features that define the boundaries of each cluster
and can then be mapped back to the input space to form the
set of clusters’ boundaries
The hypersphere is determined via the following minimizationproblem [32]: