AN APPROPRIATE FEATURE CLASSIFICATION MODEL USING KOHONEN NETWORK

Abstract

Self-Organizing Maps are widely used unsupervised neural network architecture to discover group of structures in a dataset. Feature Selection plays a major role in Machine Learning. “An Appropriate Feature Classification Model using Kohonen Network (AFCM)” is based on Recurrent Neural Network approach for feature selection which clusters relevant and irrelevant features from the dataset present in cloud environment. The proposed model not only clusters relevant and irrelevant features but also refine the clustering process by minimizing the errors and irrelevant features. The AFCM consists of Feature Selection Organizer and Convergence SOM. In the Feature Selection Organizer, features are clusters into Relevant and Irrelevant Feature classes. The Convergence SOM helps to improve the prediction accuracy in the Relevant Feature set and to reduce the irrelevant features. The efficiency of the proposed model is extensively tested upon real world medical datasets. The experimental result on standard medical dataset shows that the AFCM is better than the Traditional models.