A neural network based algorithm for complex pattern classification problems

Abstract This work presents an application of neural networks in pattern classification. A new algorithm for automatic classification of data is presented. The algorithm makes use of a competitive neural network to aid the classification process. The algorithm gets a data set D and segments it into clusters. The only prior given information is a number of auxiliary centers and a threshold distance. The algorithm uses the Mahalanobis metrics to cluster the data and find itself the number of classes. Some tests were made in artificially generated data sets with complex distributions and compared to standard classification methods that use Euclidian distance as its metrics.