Small sample learning classifies Parkinson's disease patients based on their walking behavior

Parkinson's disease (PD) is a progressive neurodegenerative disorder characterized by motor and non-motor symptoms that impair gait and daily activities. In this study, we propose a novel method based on recurrence plot and recurrence triangle (RT) patterns to classify PD patients and healthy controls. We analyze a toy model using the Rössler attractor as well as real-world walking datasets. The RT-based approach showed exceptional performance, achieving almost 100% classification accuracy on the toy model and real-world datasets. Statistical analysis revealed that specific RT patterns-type 13, type 20, and type 51 for triangle size L=4, and type 787 and type 819 for triangle size L=5-occur more frequently in healthy individuals and reflect structured, rhythmic, and adaptive gait dynamics. In contrast, triangle types 1, 60, and 64 are observed more frequently in PD patients, capturing irregular motion, fluctuations, or slow behavior. These findings demonstrate that RT patterns provide interpretable features to distinguish a healthy and pathological gait. This study highlights the potential of RT-based analysis as an accurate and interpretable tool for early detection and diagnosis of PD, paving the way for future clinical applications.