GA and PSO Based Prototype Generation for Nearest Prototype Classifiers

Abstract

K-nearest neighbor (KNN) algorithm is one the simplest and most commonly used classifications methods. The large memory and high computation cost for large training samples for KNN can be overcome by generating prototypes for each of the class using evolutionary algorithms. Nearest prototype classifiers (NPC), to some extent, resemble to the one-nearest neighbor classifier (KNN with K = 1). The aim of this study is to inspect the applications of two evolutionary algorithms namely: Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) for generating the prototypes for nearest prototype for classification of benchmark datasets. The proposed method is composed of two steps. At the first step, the prototypes representing each class are generated using real encoded GA and PSO with simple KNN using fitness function like Euclidean distance. Further, as part of the second step, the class of the test sample is determined according to the distance of the test sample to the GA and PSO generated prototypes, employing nearest prototype. To demonstrate the effectiveness of this novel method, the results of proposed method are compared with the classification results of the simple K-nearest neighbor (K = 1), prototypes identified by class means, and centroids of three clustering methods namely, K-means, farthest first and entropy based fuzzy clustering as prototypes. The performance is measured using confusion matrix. The proposed method showed enhanced classification accuracy and requires very less computational time when compared to KNN (K = 1) classifier.