Towards simultaneous prototype and Feature Generation

Nearest-neighbor (NN) methods are among the most popular and highly effective techniques used in pattern recognition tasks. However, these methods have several drawbacks that impair their performance in large scale problems and noisy data sets. Some of these disadvantages includes its high storage requirements, its sensitivity to noisy instances, and the computational cost for estimating the distance among all instances. To address these problems different techniques like Prototype Generation (PG) to reduce the number of instances, and Feature Generation (FG) to obtain a new set of features have been proposed; traditionally, both techniques have been applied separately. This paper introduces a new method for simultaneous generation of prototypes and features in order to obtain a good tradeoff between accuracy of classification with a NN classifier, instance reduction rate and feature reduction rate. The method presented is based on the algorithm NSGA-II; the main idea of the proposed method is to combine instances and attributes to produce a set of prototypes and a new feature space for each class of the classification problem via genetic programming. The proposed approach overcomes some limitations of NN without compromising its performance in classification task. Experimental results are reported and compared with several other techniques.