Databases Reference
In-Depth Information
•
Then the crossover and mutation operations are performed to get the
next candidate subset.
Generally, genetic algorithm exhibits tournament selection as the
prime operation.
52
Finally a fitness function evaluates the subset. Every
chromosome in the population represents an example of the set, where each
gene is a feature. There is not a unique representation of the features in the
chromosomes. In the case of a transactional database in database mining
and the vector space model in text mining. A gene has a value
{
}
,
meaning absence/presence of that feature in the example, respectively. The
weighting approach
53
is also generalized. Especially in text mining, where
the features are represented by their frequency in the document or another
value based on it. Nevertheless, there are not too many approaches dealing
with other types of data, where the uncertainty is considered. The rest of the
genetic parameters are not generally fixed. The size of the population does
not seem to have a relation with the number of features, and the crossover
and mutation rate utilized are the standard (0.6 and 0.001, respectively).
Most of the approaches consider Wrapper methods, although the filter ones
seem to be the most adequate in problems with a large number of features,
especially when they are combined with GAS.
54
0, l
8.3.2.
ELSA
ELSA springs forms algorithms originally motivated by artificial life models
of adaptive agents in ecological environments.
84
Modeling reproduction
in evolving populations of realistic organisms requires that selection, like
any other agent process, be locally mediated by the environments in
which the agents are situated. In a standard evolutionary algorithm, an
agent is selected for reproduction. Based on how its fitness compares to
that of other agents. In ELSA, an agent (candidate solution) may die,
reproduce, or neither based on an endogenous energy level that fluctuates
via interactions with the environment. The energy level is compared
against a constant selection threshold for reproduction. By relying on
such local selection, ELSA reduces the communication among agents to
a minimum. The competition and consequent selective pressure is driven
by the environment.
85
There are no direct comparison with other agents.
Further the local selection naturally enforces the diversity of the population
by evaluating genetic individuals based on both quality measurements and
on the number of similar individuals in the neighborhood in objective
