A Hybrid Approach of ANN with PSO for Classification Problems

An Artificial Neural Network represents a complex model which is influenced from the working of the human brain neurons. Such models have attached weights along with some bias which can be learned using loss functions and backpropagation. Its performance to a great extent depends upon the structure of the model and the various learning algorithms used to calculate and reduce its error. Backpropagation is generally used in Neural Networks to reduce the error by training the weights and making the model perform better. This work aims to show that the backpropagation algorithm can be replaced by the Particle Swarm Optimization algorithm to train neural networks and is shown to give a better accuracy on such models when compared with the models trained using backpropagation. Such an Artificial Neural Network is employed on a number of benchmark datasets from UCI, Machine Learning Repository to achieve this goal.This approach achieves a perfect test accuracy for most of the datasets considered, surpassing all the previous test accuracies ever found and thus clearly reflecting the effectiveness of this approach in Neural Networks.