Optimizing hyperparameters is essential for improving the performance of feedforward neural networks and plays a critical role in designing an effective classification model. Key hyperparameters include the number of hidden layers, the number of neurons in each layer, the activation function, learning rate, and momentum. Metaheuristic algorithms have been widely used to optimize the hyperparameters of the multilayer perceptron (MLP). In the literature. In this paper, we propose a metaphor-less optimization algorithm called MLP-Rao, which utilizes Rao algorithms to optimize the MLP architecture for data classification tasks. The Rao algorithms are simple population-based optimizer that do not rely on metaphor of any creature concept. Like Jaya metaheuristic algorithm, Rao algorithms are free from control parameters and have demonstrated better performance compared to metaphor-based optimizers. Additionally, an efficient fitness function is utilized to measure the fitness of the solutions in a population. This fitness function maintains a balance between classification performance and architecture complexity in order to avoid model’s overfitting risks. The proposed approach is validated on standard benchmark medical datasets. Comprehensive experiments reveal that our method achieves promising results compared to state-of-the-art methods. In addition, the Rao algorithms require less computational time compared to metaphor-based optimization methods. These results emphasize the efficacy of the proposed approach in optimizing MLP hyperparameters and suggesting it as an alternative approach to metaphor-based optimization algorithms.