Activity recognition in patients with tremor: Integrating time-length windows for enhanced detection

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Machine Learning in Healthcare

Background:

• Pathological tremor, common in essential tremor and Parkinson's disease, severely impacts daily life.
• Current tremor assessment methods are subjective and lack continuous monitoring capabilities.
• Existing activity recognition studies often exclude tremor patients or use suboptimal fixed window sizes.

Purpose of the Study:

• To develop a novel method for classifying activities of daily living (ADLs) using a single smartwatch for improved tremor monitoring.
• To investigate the effectiveness of various windowing techniques and machine learning models for ADL classification in tremor patients.
• To provide a more objective and comprehensive evaluation of tremor and its impact on daily activities.

Main Methods:

• Utilized time series data from a single smartwatch to classify ADLs.
• Processed data using various windowing techniques, including integrated time-length windows.
• Applied machine learning models: Support Vector Machines, Random Forest, and Extreme Gradient Boosting.
• Evaluated model performance using accuracy, precision, recall, and F1-score.

Main Results:

• Demonstrated significant improvements in activity recognition accuracy.
• Highlighted the effectiveness of integrated time-length windows for feature extraction.
• Showcased the potential of smartwatch-based ADL classification for tremor monitoring.

Conclusions:

• The proposed smartwatch-based method offers a more objective and comprehensive approach to tremor evaluation.
• This technology can enhance treatment strategies by linking tremor to specific daily activities.
• Future research can build upon this to refine tremor assessment and patient care.