Swarm Intelligence Approach for Optimizing Neural Network Architectures in Medical Data Classification

The capability of an Artificial Neural Network (ANN) to achieve accurate classification outcomes depends heavily on its architectural design, particularly the number of hidden layers, nodes, and other hyperparameters. Selecting an optimal architecture remains a crucial yet challenging task for improving network accuracy, especially in medical informatics, where reliability is essential. This paper introduces a novel model, MRFO-MLP, which integrates the Manta Ray Foraging Optimization (MRFO) algorithm with a Multi-layer Perceptron (MLP). MRFO is a swarm-intelligence metaheuristic inspired by the cooperative foraging behaviour of manta rays and is employed to explore potential MLP configurations within the solution space. In the MRFO-MLP model, each candidate solution is represented as a binary vector, which reduces both computational time and implementation complexity. Furthermore, a customised fitness function is designed to evaluate candidate architectures by simultaneously considering network complexity and generalisation error, thereby mitigating the risk of overfitting. The proposed model is evaluated using five medical datasets from the UCI repository. Its performance is compared with several optimisation algorithms, including the Genetic Algorithm (GA), Particle Swarm Optimisation (PSO), and traditional machine learning classifiers. Experimental results demonstrate that MRFO-MLP produces efficient and accurate MLP architectures, achieving the best performance on three datasets and the second-best performance on another, highlighting its competitive effectiveness for medical data classification tasks.