Review of current Online Dynamic Unsupervised Feed Forward Neural Network classification
Roya Asadi, Haitham Sabah Hasan, Sameem Abdul Kareem · 2014
Online Dynamic Unsupervised Feed Forward Neural Network (ODUFFNN) classification is suitable to be applied in different research areas and environments such as email logs, networks, credit card transactions, astronomy and satellite communications. Currently, there are a few strong methods as ODUFFNN classification, although they have general problems. The goal of this research is an investigation of the critical problems and comparison of current ODUFFNN classification. For experimental results, Evolving Self-Organizing Map (ESOM) and Dynamic Self-Organizing Map (DSOM) as strong related methods are considered; and also they applied some difficult datasets for clustering from the UCI Dataset Repository. The results of the ESOM and the DSOM methods are compared with the results of some related clustering methods. The clustering time is measured by the number of epochs and CPU time usage. The clustering accuracies of methods are measured by employing F-measure through an average of three times performances of clustering methods. The memory usage and complexity are measured by the number of input values, training iterations, clusters; and densities of clusters. (Abstract) Keywords—Neural Network (NN) model, Feed Forward Unsupervised Classification, Training, Epoch, Online Dynamic Unsupervised Feed Forward Neural Network (ODUFFNN) (key words)