Training neural networks using Salp Swarm Algorithm for pattern classification

Pattern classification is one of the popular applications of neural networks. However, training the neural networks is the most essential phase. Traditional training algorithms (e.g. Back-propagation algorithm) have some drawbacks such as falling into the local minima and slow convergence rate. Therefore, optimization algorithms are employed to overcome these issues. Salp Swarm Algorithm (SSA) is a recent and novel nature-inspired optimization algorithm that proved a good performance in solving many optimization problems. This paper proposes the use of SSA to optimize the weights coefficients for the neural networks in order to perform pattern classification. The merits of the proposed method are validated using a set of well-known classification problems and compared against rival optimization algorithms. The obtained results show that the proposed method performs better than or on par with other methods in terms of classification accuracy and sum squared errors.