Training Multi-Layer Perceptron with Enhanced Brain Storm Optimization Metaheuristics

In the domain of artificial neural networks, learning process represents one most challenging tasks. Since classification accuracy highly depends on weights and biases, it is crucial to find its optimal or suboptimal values for problem at hand. However, a very large search space, difficult proper connection biases. Employing traditional optimization algorithms this issue leads slow convergence prone get stuck in local optima. Most commonly, back-propagation used multi-layer-perceptron training can lead vanishing gradient issue. As an alternative approach, stochastic algorithms, such as nature-inspired metaheuristics are more reliable complex tax, finding biases network training. work, we propose enhanced brain storm optimization-based algorithm networks. simulations, ten binary benchmark datasets with different difficulty levels evaluate efficiency proposed algorithm. The results show that approach promising achieved better than other state-of-the-art approaches majority terms speed, due capability balancing intensification diversification avoiding minima. obtained best eight out observed dataset, outperforming all by 1–2% average. When mean observed, dominated nine datasets.