Analysis of Variable Learning Rate Back Propagation with Cuckoo Search Algorithm for Data Classification

For the data classification task back propagation (BP) is the most common used model to trained artificial neural network (ANN). Various parameters were used to enhance the learning process of this network. However, the conventional algorithms have some weakness, during training. The error function of this algorithm is not explicit to locate the global minimum, while gradient descent may cause slow learning rate and get stuck in local minima. As a solution, nature inspired cuckoo search algorithms provide derived free solution to optimize composite problems. This paper proposed a novel meta-heuristic search algorithm, called cuckoo search (CS), with variable learning rate to train the network. The proposed variable learning rate with cuckoo search algorithm speed up the slow convergence and solve the local minima problem of the backpropagation algorithm. The proposed CS variable learning rate BP algorithms are compared with traditional algorithms. Particularly, diabetes and cancer benchmark classification problems datasets are used. From the analyses results it show that proposed algorithm shows high efficiency and enhanced performance of the BP algorithm.