Saeide Sheikhpour - Faculty of Electrical Engineering, University of Birjand, Iran
Mahdieh Sabouri
Seyed-Hamid Zahiri

Tuning optimum parameter of neural networks, such as weights and biases, has major effects on their performance improvement. Estimation of optimum values for these parameters requires strong and effective training methods, so that the error of the training data reaches its minimum. This paper presents, a suitable training method for optimizing neural networks parameters using a novel hybrid GA-GSA algorithm. Extensive experimental results on different benchmarks show that the hybrid algorithm, performs equal to or better than standard GSA, and backpropagation algorithm

neural network, gravitational search algorithm, genetic algorithm, back propagation algorithm