A novel hybrid multilayer perceptron neural network with improved grey wolf optimizer

The multilayer perceptron (MLP), a type of feed-forward neural network, is widely used in various artificial intelligence problems in the literature. Backpropagation is the most common learning method used in MLPs. The gradient-based backpropagation method, which is one of the classical methods, has some disadvantages such as entrapment in local minima, convergence speed, and initialization sensitivity. To eliminate or minimize these disadvantages, there are many studies in the literature that use metaheuristic optimization methods instead of classical methods. These methods are constantly being developed. One of these is an improved grey wolf optimizer (IMP-GWO) proposed to eliminate the disadvantages of the grey wolf optimizer (GWO), which suffers from a lack of search agent diversity, premature convergence, and imbalance between exploitation and exploration. In this study, a new hybrid method, IMP-GWO-MLP, machine learning method was designed for the first time by combining IMP-GWO and MLP. IMP-GWO was used to determine the weight and bias values, which are the most challenging parts of the MLP training phase. The proposed IMP-GWO-MLP was applied to 20 datasets consisting of three different approximations, eight regression problems, and nine classification problems. The results obtained have been suggested in the literature and compared with the gradient descent-based MLP, commonly used GWO, particle swarm optimization, whale optimization algorithm, ant lion algorithm, and genetic algorithm-based MLP methods. The experimental results show that the proposed method is superior to other state-of-the-art methods in the literature. In addition, it is thought that the proposed method can be modeled with high success in real-world problems.