Hybrid kernel density estimation for discriminant analysis with information complexity and genetic algorithm

A new hybrid approach is proposed which is computationally effective and easy to use in selecting the best subset of predictor variables in discriminant analysis (DA) under the assumption that data sets do not follow the normal distribution. The proposed approach integrates kernel density estimation for discriminant analysis (KDE-DA) and the information theoretic measure of complexity (ICOMP) with the genetic algorithm (GA). The ICOMP plays an important role in finding both the best bandwidth matrix for KDE-DA and the best subset of predictor variables which discriminate between the groups. The genetic algorithm (GA) is introduced and used within KDE-DA as a clever stochastic search algorithm. To show the working of this new and novel approach, six benchmark real data sets are considered and the results are compared with results of linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), and k-nearest neighbor discriminant analysis (k-NNDA) to choose the best fitting model. The experimental results show that the proposed hybrid kernel density estimation approach outperforms LDA, QDA, and k-NNDA.