Extended Self-Organizing Map With Ubiquitous Counter Propagation Network In Classification For Diabetic Database

Authors

  • S. Sutha, N. Gnanambigai, P. Dinadayalan

Abstract

Recently, Intelligent based neural networks are proven to be a promising method for data processing and identify the rupturing part by using their powerful non-linear modeling capability. The artificial neural networks are a structure and function of the human brain used in machine learning. It incorporates unsupervised with a supervised algorithm which enhances the Data classification by specifying some rules to categorize the data into various categories. Most popular techniques to cluster and visualize the data in areas of science are Self-organizing maps (SOMs). The more appropriate framework is counter-propagation (CPN) has been vividly accomplished in various platform such as statistical analysis, pattern classification, function approximation. This simple network structure has a less error prone by a series convergence outstanding to its collaboration of the Kohonen self-organizing map and classification network model. This paper projects a hybrid SOM (E-SOM) with Decision tree that modifies with the ubiquitous counter propagation network (U-CPN) model in classification for the Diabetic database. These are three-layer network architectures that implement the rules for learning. Whereas, network architecture consists of an input layer, a layer of Kohonen and a layer of Grossberg. The extended self-organizing map model, however, applies a changed learning rule from Kohonen to train both the Kohonen layer and the output layer. The results of the simulation provide optimal learning by applying extended self-organizing map model, the enhanced counter propagation model with minimal performance errors. The classifier eliminates the classification errors and results are obtained from diabetic database which indicates that the extended U-CPN has an enhanced classification speed and a much higher classification precision than the existing conventional method.

Published

2020-12-01

Issue

Section

Articles