MLP-FOX: A Metaheuristic Approach-Based Training Multilayer Perceptron for Data Classification

Training a Multilayer Perceptron (MLP) using metaheuristic optimization algorithms serves as a benchmark problem for demonstrating the effectiveness of emerging metaheuristic optimizers. In this study, we explore the efficacy of a novel metaheuristic optimizer namely, Fox algorithm, to optimize the weights and biases of an MLP. However, backpropagation algorithm is a traditional and well-known gradient-based optimizer is commonly used in training MLP, may become trap in local minima in many cases due to complex multimodal training error curves. This issue can diminish the classification accuracy. To overcome this problem, stochastic optimization algorithms play a vital role in improving error convergence and generalization capability of the MLP. In our work, exhaustive experimentation and analysis has been performed using five standard classification medical datasets to show the effectiveness of the Fox algorithm. The experimental results compare with backpropagation (generalized delta rule) and three recently proposed metaheuristic optimizers to evaluate its capability to converge to optimal solutions and enhance the overall performance of the MLP. Based on error convergence rate, accuracy, and F1-score, the proposed approach reveals promising classification performance compared to competitors.