Parkinson’s disease is a neuro-degenerative disorder characterized by different motor symptoms, including several gait impairments. Gait analysis is a suitable tool to support the diagnosis and to monitor the state of the disease. This study proposes the use of non-linear dynamics features extracted from gait signals obtained from inertial sensors for the automatic detection of the disease. We classify two groups of healthy controls (Elderly and Young) and Parkinson’s patients with several classifiers. Accuracies ranging from 86% to 92% are obtained, depending on the age of the healthy control subjects.