Pattern classification using modified enhanced fuzzy min-max neural network

In this paper modified enhanced fuzzy min max (modified-EFMMN) has been proposed for pattern classification. The objectives of modified-EFMM are firstly, to lift the classification accuracy, secondly to reduce the network complexity and thirdly to utilize minimum number of features to provide classification decision. The modified-EFMM handles overlap among the different class hyperbox more stringently, it gives the more classification accuracy. The pruning method is applied to reduce network complexity by removing unwanted hyperboxes based on confidence factor and user defined threshold value. Then closed hyperboxes are converted to open hyperboxes and the open hyperboxes with minimum number of features that contribute more to the accuracy are selected. The performance of modified-EFMM is evaluated by using benchmark Iris dataset taken from UCI machine learning repository. The obtained conclusion shows that the modified-EFMM gives more accuracy with less number of features and also with less number of hyperboxes.