Published Paper Details:

Artificial Neural Networks (ANNs) are the digitized model of the human brain and can be effectively used in the context where it is very difficult for humans to detect the underlying complex non-linear relationships in the data. In this paper, a feed forward single hidden layer neural network is trained using Back propagation algorithm for data classification. In real world, patterns are characterized by many features and not all features are relevant for a particular task and using all the features for training an ANN often reduces the learning rate. Hence, to increase the learning speed of the ANN, the most popular Rough Set Theory (RST) algorithms have been used to determine the most significant features and then training ANN only on the relevant features reduce the time complexity of the Neural Networks as well as helps in increasing the generalization ability. The trained ANN is 10-fold cross validated by conducting experiments on the datasets taken from UCI ML repository and the results revealed that the removal of superfluous features in the training data enhances the performance of ANN in increasing the learning speed and classification accuracy.