SVM Classification System
The hybrid of the Genetic and Support Vector Machine consists of two key elements, SVM and GA classifiers. The primary role of the GA is to identify subsets of features while SVM evaluates the subsets during the process of classification (Pustejovsky and Stubbs, 2012). After the classification, the ‘best result’ undergoes a further fitness evaluation, to come up with the most suitable among the best. At every stage of evaluation, the computation of accuracy and consistent training of SVM is done. The fittest chromosomes are then ushered to the next phase of computations to determine the fittest among them. For instance, let us suppose η is the number features available for representing the data available for classification. The chromosome is then represented by the binary vector of η dimensions. Therefore, in the event, the selected bit is 1, what this means is that the feature is selected, but if the bit is 0 it has not been selected. The following equation represents the number of feature subsets based on the chromosome representation.
ηc = 2 η
Where ηc represents the number of feature subsets, η represents the number of features.
On the other hand, the fitness function has to pick different methods as a standard for the function. The methods may include, the cost of performing classification or the accuracy of the classification or in some cases both (Koonin & Galperin, 2013). The individual fitness function is calculated using a training set by evaluating the SVM. Henceforth, the fitness function used in the research containing classification accuracy is as seen below
fitness (γ) = accuracy(γ)
where accuracy(γ) represents the LOOCV (leave one out cross validation) accuracy of the classier with feature subset selection γ.
According to Cagnoni, and Gustavo (2008), the GA a set of possible solutions and comes up with the best possible solution after a series of repetitive computations. It evaluates each individual’s fitness regarding quality ratio through a fitness function. The GA comes up with the best subset of training and testing sets.