A research team led by Professor Jungpil Shin of the School of Computer Science and Engineering at the University of Aizu has published a new study on handwriting-based decision support for Parkinson's disease (PD) using machine learning. The research article, entitled "HandPD37: advanced handwriting dataset for Parkinson's disease diagnosis using stack ensemble ML," was published in the international journal Scientific Reports*. 
* Scientific Reports  https://www.nature.com/articles/s41598-026-63031-y

Parkinson's disease is a neurodegenerative disorder associated with motor symptoms such as tremor, bradykinesia, and rigidity. These symptoms can also affect fine motor activities such as handwriting. In this study, the researchers developed a new handwriting dataset, named HandPD37, with the aim of quantitatively capturing subtle changes in handwriting and exploring their potential use in supporting the assessment of Parkinson's disease.

The HandPD37 dataset contains handwriting data collected from 132 participants, including 78 patients with Parkinson's disease and 54 healthy controls. Participants completed 37 different handwriting and drawing tasks on a digital tablet, including repetitive character writing, circles and spirals, Japanese kana and kanji characters, wave patterns, and geometric shapes. The tablet recorded detailed time-series information such as X-Y coordinates, pen pressure, timestamps, pen orientation, and tilt. 

From these data, the research team extracted 171 handwriting features, including writing speed, acceleration, jerk, pen pressure, and stroke-related characteristics. The study then applied a stacking ensemble framework combining 10 machine learning algorithms, including Random Forest, Support Vector Machine, LightGBM, and XGBoost, with Logistic Regression used as the meta-classifier. 

The experimental results showed high classification performance across the 37 handwriting tasks. When all 171 features were used, the average classification accuracy across the tasks reached 97.10%. The ensemble model also achieved the highest accuracy in 16 of the 37 tasks and generally demonstrated more stable performance than most individual machine learning models.

Several handwriting tasks were found to be particularly informative for distinguishing participants with Parkinson's disease from healthy controls. These included spiral drawing (Task 18), tracing a complex Japanese character (Task 28), wave drawing (Task 33), and a task combining Archimedean spiral drawing with repeated cursive "l" writing (Task 34). Among them, Task 18 was consistently highly ranked across multiple machine learning models, indicating that spiral drawing may provide particularly useful information for handwriting-based Parkinson's disease assessment.

The findings demonstrate the potential of tablet-based handwriting analysis as a simple, non-invasive, and quantitative tool for supporting the assessment of Parkinson's disease. Rather than replacing clinical diagnosis by medical specialists, the proposed approach may in the future provide additional objective information during routine examinations, rehabilitation assessment, or follow-up monitoring.

Future work will focus on validating the approach using larger multi-center clinical datasets and integrating additional sensor modalities, such as accelerometers, gyroscopes, and image-based handwriting features, with the aim of further improving the robustness and clinical applicability of Parkinson's disease assessment.

Publication Information
Title: HandPD37: advanced handwriting dataset for Parkinson's disease diagnosis using stack ensemble ML
Journal: Scientific Reports
Volume: 16
Article number: 30515 (2026)
DOI: 10.1038/s41598-026-63031-y 
Authors: Jungpil Shin, Abu Saleh Musa Miah, Akane Hashimoto, Koki Hirooka, Md. Al Mehedi Hasan, Najmul Hassan, Mitsuki Aihara, Ryoma Takahashi, and Shunsuke Kobayashi. The author affiliations include the University of Aizu, Hokkaido University, CAC Corporation, Dokkyo Medical University, and Teikyo University.
Published: 01 October 2026
Link to paper: https://u-aizu.ac.jp/files/Shin_et_al-2026-Scientific_Reports.pdf

Related Links

For inquiries regardingthe contents of this page.

Planning and Collaboration Division
cl-planpr@u-aizu.ac.jp