A Modified Multi-Swarm Optimization for Weights and Biases of a Multi-Layer Perceptron for Medical Data Classification

The significant increment in the amount of data managed when medical diagnosis are performed makes the rising of new technics necessary for supporting medical decisions. Artificial intelligence methods have taken relevance in the latest years. Particularly, the Multi-Layer Perceptron stands out in the medical field, reaching highlighted results when applied to different clinical disciplines. Nevertheless, it can obtain a poor performance due to the weaknesses of its learning method, known as the Back-Propagation Algorithm. It is why alternatives started to be raised, especially those based on metaheuristics. This work proposes a metaheuristic called Modified Multi-Swarm Optimization Algorithm as an alternative to the Back-Propagation Algorithm. Results using five medical datasets show that our proposal can offer a remarkable performance compared to other state-of-the-art algorithms, attaining minimal mean-square error values at lower time consumption and being able to improve accuracy, sensitivity and specificity.