Abstract
Various algorithms for supervised classification of multi-dimensional data have been implemented in the past decades. Among these algorithms, neural and statistical classifiers are two major methodologies used in the literature. In this paper, a comparison of different neural networks and statistical algorithms used for classification is presented. Three types of neural classifiers are considered: Back-propagation (BP) network, Radial Basis Function (RBF), and Learning Vector Quantization (LVQ). Meanwhile, the k-nearest neighbor (KNN) statistical classifier and linear discriminant analysis (LDA) are also discussed in order to compare the accuracy of classification with those of using neural network models. This paper includes an introduction to the theoretical background of the classifiers, their implementation procedures, and two case studies to evaluate their performance in diagnosis of diseases and glass identification. Both neural networks and statistical models are demonstrated to be efficient and effective methods for multi-dimensional data classification. For neural classifiers, the type of neural network, the type of data, and the parameters of the neural network may have considerable impact on the classification accuracy. For statistical models, the type of data distribution and the criteria used to determine the thresholds are the two most important factors in classification. In the cases studied in this paper, the overall performance of neural networks is better than that of statistical models. Finally, the comparison and discussion of these approaches are presented in view of practical and theoretical considerations.