Muhammad Izzuddin Mahali, Cries Avian, Nur Achmad Sulistyo Putro, Setya Widyawan Prakosa, Jenq-Shiou Leu
Parkinson’s disease (PD) is a prevalent neurodegenerative disorder that affects motor function and impacts millions worldwide. Early diagnosis and precise staging of PD are critical for effective management and timely intervention. Traditional gait analysis focuses on lower limb movement to differentiate between moderate and advanced PD stages; however, integrating upper and lower limb dynamics offers a more comprehensive approach to distinguishing early-stage PD from other. This study introduces a novel spatio-temporal model that integrates CNN, Transformer, and BiLSTM layers to classify PD based on the Hoehn and Yahr (H&Y) score. The architecture captures both spatial and temporal dependencies in gait patterns, with a CNN serving as the primary spatial feature extractor, utilizing a temporal-focused kernel, and augmented by a Transformer to enhance temporal feature learning. Additionally, a BiLSTM module synthesizes hierarchical spatio-temporal features through multi-path integration of CNN and Transformer outputs to model complex gait dynamics. For data augmentation, sliding windows of 500, 800, and 1200 samples were employed, and the model’s performance was rigorously assessed through 3-, 5- and 10-fold cross-subject validation. Metrics included accuracy, F1-score, recall, precision, and a voting-based confidence score, while t-SNE visualization provided insights into spatio-temporal feature differentiation. The model demonstrated high performance, achieving optimal accuracy of 0.90 and an average confidence score of 0.97 with a 1200-sample window in 5-fold validation. These findings underscore the potential of spatio-temporal deep learning architectures to advance multi-class PD classification and highlight their effectiveness in detecting early-stage PD from gait pattern signals. © The Author(s) 2026.
Department of Electronic and Computer Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan; Department of Electronics and Informatics Engineering Education, Faculty of Engineering, Universitas Negeri Yogyakarta, Yogyakarta, Indonesia; Department Electrical Engineering, Faculty of Engineering, Universitas Brawijaya, Malang, Indonesia; Department of Computer Science and Electronics, Faculty of Mathematics and Natural Sciences, Universitas Gadjah Mada, Yogyakarta, Indonesia; Department of Electrical Engineering, Faculty of Engineering, University of Jember, Jember, Indonesia
