Present paper extends the novel learning technique of evolutionary
support vector machines to $k$-class classification, $k > 2$, by
means of hybridization of the inherent evolutionary algorithm with
different multi-class support vector machines techniques. The aim
of the new paradigm is to evolve, in a simpler and direct manner,
the coefficients of the separating surfaces induced by support
vector machines. We propose three methods to be integrated in the
new approach, i.e. one-against-one, one-against-all and decision
directed acyclic graph. Comparison is achieved on the benchmark
data set of Fisher's Iris flowers and proves that, for low
dimensional problems, the three approaches perform in a similar
fashion and are competitive to their canonical counterparts.