|
Abstract:
Support vector machine (SVM) has achieved excellent performances in the classification of hyper spectral data and wide variety of applications. Nevertheless, how to effectively reduce the complexity features of training dataset for SVM is still a serious challenge. In this paper, an efficient scheme of differential evolution (DE)-based dimensionality reduction approach is proposed to use a simple searching criterion function, called adaptive estimated nearest neighbors (AENN), to optimize the reduction of noisy instances in a classification process for the SVM. With such an efficient criterion, DE algorithm can find a global optimal solution for the AENN and SVM kernel parameters to improve the classification accuracy. Several UCI benchmark datasets are considered to compare the proposed hybrid DE-AENN dimensionality reduction strategy with the previously published methods. Experimental results show that the proposed hybrid framework is capable of achieving better performance than other existing methods and is feasible to construct a condensed nearest neighbors of training dataset to enhance the classification accuracy for SVM.
|
