1. Artificial neural network based classification and performance evaluation using benchmark datasets
Authors : Josephina Paul
Pages : 9-16
DOI : http://dx.doi.org/10.21172/1.162.02
Keywords : Artificial neural network, Backpropagation algorithm, classification, confusion matrix, accuracy Abstract :Classification is the fundamental task for many machine learning applications. Therefore, it is highly demanding and active research area presently as it has been for the past many decades. Since Artificial Neural Network supports generalization and robust in solving non linear problems such as multiclass classification, we choose Feed-forward neural network with backpropagation in our experiments. In this study, we classify the benchmark datasets, the Iris and E.coli datasets, which are typical classification problems as they contain multiple classes. By several trials we arrive at an optimal network architecture 4-10-3 for Iris dataset and 7-15-8 for E.coli dataset. Secondly, the success of neural network is mainly depends on its training and hence, we have selected two learning algorithms having faster convergence, ‘trainlm’ and ‘traincgb’ from Matlab for training the network. The training, validation and testing of the datasets are done in the ratio 70:15:15. The over all accuracy obtained are 98% and 99.3% for the Iris dataset, and 89.9% and 91.7% for E.coli dataset with, ‘trainlm’ and ‘traincgb’ respectively. The accuracies achieved are also compared against those obtained for standard algorithms in the literature and found superior to them.
Citing this Journal Article :Josephina Paul, "Artificial neural network based classification and performance evaluation using benchmark datasets", Volume 16 Issue 2 - April 2020, 9-16
Click here to Submit Copyright Takedown Notice for this article.