Hidden Markov Model-Driven Dynamic Walking Segment Analysis for Parkinson’s Disease Classification

• Sungwook Hur,
• Jieming Zhang,
• Moon-Hyun Kim,
• Tai-Myoung Chung,

Vol. 19, No. 7, July 31, 2025
10.3837/tiis.2025.07.006, Download Paper (Free):

• Ensemble learning
• Hidden Markov Model
• Machine Learning
• Parkinson disease
• Time Series Data

Abstract

Parkinson’s Disease (PD) is a progressive neurodegenerative disorder characterized by abnormal gait patterns, which can be captured through wearable accelerometers. However, accurately extracting walking segments from noisy accelerometer data remains challenging, and conventional threshold-based segmentation methods often fail to capture the subtle yet critical dynamics of gait. To address these limitations, we propose a Hidden Markov Model (HMM)-driven framework for dynamic walking segment analysis, coupled with an optimized AdaBoost classification approach. The HMM-driven segmentation algorithm models the underlying gait states to reliably isolate high-quality walking segments, while the optimized AdaBoost classifier leverages variance-based weight initialization to emphasize informative samples. By integrating these components, our method enhances the robustness and discriminative power of PD classification. We evaluated on the publicly available PD-BioStampRC21 dataset. The proposed approach achieved 93% classification accuracy, with a sensitivity of 1.0, an AUC of 0.98 and Matthews Correlation Coefficient of 0.88, significantly outperforming conventional threshold-based segmentation and standard ensemble learning methods. These results demonstrate that the integration of HMM-driven segmentation with optimized ensemble learning substantially improves PD classification accuracy, offering promising implications for clinical applications in automated PD assessment and monitoring.

Statistics

Cite this article