Feature and instance selection through discriminant analysis criteria

Feature selection and instance are two data preprocessing methods widely used in mining pattern recognition. The main goal is to reduce the computational cost of many learning tasks. Recently, joint feature has been approached by solving some global optimization problems using meta-heuristics. This approach not only computationally expensive, but also does exploit fact that usually have a structured manifold implicitly hidden its labels. In this paper, we address scores derived from discriminant analysis theory. We present three approaches for selection. first scheme wrapper technique, while other schemes filtering techniques. approaches, search process uses genetic algorithm where evaluation criterion mainly given score. score depends simultaneously on subset candidate best corresponding instances. Thus, instances determined finding performance proposed quantified studied image classification with Nearest Neighbor Support Vector Machine Classifiers. Experiments conducted five public datasets. compare our several state-of-the-art methods. experiments performed show superiority over baseline