A Survey on Baysean Classification
- 2-D example of two-straight lines
(39 pages)
Let me summarize it as
A candidate solution is called a non-dominated iff there is no ohter better
solution w.r.t all the objectives.
Further more [1] refer to [3] as for "how to rank the individual" as follows.
As there is no single objective function to determine the
fitness of the different individuals in a Pareto optimization,
the ranking scheme presented by Fonseca and Fleming is employed,
and the "degree of dominance" in attribute
space is used to rank the population. Each individual is given a rank based on
the number of individuals in the population that are preferred to it, i.e.
for each indi-vidual the algorithm loops through the whole population counting the number of
preferred individuals. "Preferred to" is implemented in a strict Pareto sense, but one
could also combine Pareto optimality with the satisfaction of objective goal levels, as
discussed in Fonseca and Fleming. To be more in detail, see:
Sequential Niching