Identificação da doença de Parkinson pela dinâmica da marcha sob dupla tarefa com machine learning

Abstract

This study examines the application of machine learning for Parkinson's Disease classification using gait signal analysis under dual-task cognitive conditions. The methodology uses vertical ground reaction force time series from sensors attached to 27 individuals during simultaneous walking and serial mathematical subtraction. Data are preprocessed and segmented to extract global features that feed a Multilayer Perceptron (MLP) neural network, evaluated by subject-stratified cross-validation with SMOTE balancing. The model achieves 96% mean accuracy and 100% sensitivity, correctly identifying all patients with the condition without any false negatives. Results support the clinical viability of the dual-task approach and its potential for deployment in low-cost computational devices for Parkinson's Disease screening.