Aplicação de Redes Perceptron Multicamadas no Diagnóstico da Doença de Alzheimer com IA Explicável

Abstract

This article presents the development of a Multi-Layer Perceptron (MLP) neural network for the binary classification of Alzheimer’s disease (AD) using structured clinical data. The goal is to offer a low-cost, high-sensitivity alternative for initial screening, overcoming the socioeconomic barriers associated with imaging tests. The methodology involved the preprocessing of 2,149 records, including normalization and class balancing, as well as an optimized architecture with Batch Normalization, Dropout, and L2 regularization. The results demonstrated an accuracy of 79.07\% and a recall of 87.5\%. It is concluded that the model is effective in minimizing false negatives in scenarios with limited infrastructure.