Evolutionary Creation of Discriminant Functions Based On Margin Maximization Method

摘要

There are many approaches on pattern recognition with Genetic Algorithms (GAs). Here we propose a method of evolving discriminant function with the GA and a margin maximization method. We assume that discriminant functions can be approximated by a linear combination of functions that are products sum of power of input variables. We use the GA to search the exponentials of each input variable of terms in discriminant fucntions and the coefficients of terms are caculated by the margin maximizaiton method. Simulations are performed on several datasets in UCI Repository. We obtain that the proposed method is effective in improving the classification performance.