Databases Reference
In-Depth Information
Fig. 8.6.
Structure of a chromosome (bit set).
GA chromosomes. SVMs use kernel
68
functions to transform input features
from lower to higher dimensions. Implementation of GAs is achieved by
translating the parameters into a coded string of binary digits, as is done
in this proposed hybrid. These strings denote the attributes present in the
data sets, with the length of the string being equal to the
N
+1, where
N
is the number of attributes excluding the class attribute. A typical
structure (a chromosome) is illustrated in following Fig. 8.6. After each
generation, the algorithm would then check two termination criteria.
Firstly, if convergence is achieved — the case when all chromosomes in
the population possess the same fitness levels — the evolution process
can then be halted. The maximum number of generations that the user
permits the algorithm to run before stopping the process is set prior to
commencement. The second criterion is based on this parameter that is
decided by the user. If convergence is not reached before the maximum
number of generations,
66,69
the algorithm will cease.
Comparison with pure SVM:
The GA-SVM hybrid was tested with
pure SVM to investigate the performance of the additional attribute
selection component. Pure SVM in this case means that no attribute
selection was done on the data sets. The GA-SVM hybrid incorporates the
stochastic nature of genetic algorithms together with the vast capability of
support vector machines in the search for an optimal set of attributes. The
eradication of the redundant attributes using the GA-SVM hybrid improves
the quality of the data sets and enables better classification of future unseen
data.
8.4.1.
Neuro-Fuzzy feature selection
This feature selection approach is shown to yield a diverse population of
alternative feature subsets with various accuracy/complexity trade-off. The
algorithm is applied to select features for performing classification with
fuzzy model. Fuzzy
89
models involving only a few inputs can be more
compact and transparent, thus offering improved interpretability of the
fuzzy rule base. Such subtleties are often overlooked when feature selection
