Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-dataset Multivariate Time-series Model for Parkinson's Diagnosis via Keyboard Dynamics

Arianna Francesconi, Donato Cappetta, Fabio Rebecchi, Paolo Soda, Valerio Guarrasi, Rosa Sicilia

TL;DR

This paper tackles the challenge of early Parkinson's diagnosis by leveraging keystroke dynamics as a remote, scalable biomarker. It proposes a four-stage pipeline—pre-processing, pre-training on large fixed-text KD datasets, targeted fine-tuning on a medium free-text dataset, and external validation on an independent cohort—to achieve robust cross-dataset generalization. By evaluating eight deep-learning time-series architectures and three data-balancing strategies (notably IMBALMED), the study demonstrates that hybrid CNN-RNN and transformer-based models can exceed 90% AUC-ROC with F1-scores above 70% in external validation, with a temporal convolutional network achieving 91.14% AUC-ROC. The findings highlight the potential of KD-based digital biomarkers for telemonitoring and early PD screening, while underscoring the importance of cross-dataset validation and balanced training; future work will explore multimodal inputs and explainability to enhance clinical adoption.

Abstract

Parkinson's disease (PD) presents a growing global challenge, affecting over 10 million individuals, with prevalence expected to double by 2040. Early diagnosis remains difficult due to the late emergence of motor symptoms and limitations of traditional clinical assessments. In this study, we propose a novel pipeline that leverages keystroke dynamics as a non-invasive and scalable biomarker for remote PD screening and telemonitoring. Our methodology involves three main stages: (i) preprocessing of data from four distinct datasets, extracting four temporal signals and addressing class imbalance through the comparison of three methods; (ii) pre-training eight state-of-the-art deep-learning architectures on the two largest datasets, optimizing temporal windowing, stride, and other hyperparameters; (iii) fine-tuning on an intermediate-sized dataset and performing external validation on a fourth, independent cohort. Our results demonstrate that hybrid convolutional-recurrent and transformer-based models achieve strong external validation performance, with AUC-ROC scores exceeding 90% and F1-Score over 70%. Notably, a temporal convolutional model attains an AUC-ROC of 91.14% in external validation, outperforming existing methods that rely solely on internal validation. These findings underscore the potential of keystroke dynamics as a reliable digital biomarker for PD, offering a promising avenue for early detection and continuous monitoring.

Cross-dataset Multivariate Time-series Model for Parkinson's Diagnosis via Keyboard Dynamics

TL;DR

This paper tackles the challenge of early Parkinson's diagnosis by leveraging keystroke dynamics as a remote, scalable biomarker. It proposes a four-stage pipeline—pre-processing, pre-training on large fixed-text KD datasets, targeted fine-tuning on a medium free-text dataset, and external validation on an independent cohort—to achieve robust cross-dataset generalization. By evaluating eight deep-learning time-series architectures and three data-balancing strategies (notably IMBALMED), the study demonstrates that hybrid CNN-RNN and transformer-based models can exceed 90% AUC-ROC with F1-scores above 70% in external validation, with a temporal convolutional network achieving 91.14% AUC-ROC. The findings highlight the potential of KD-based digital biomarkers for telemonitoring and early PD screening, while underscoring the importance of cross-dataset validation and balanced training; future work will explore multimodal inputs and explainability to enhance clinical adoption.

Abstract

Parkinson's disease (PD) presents a growing global challenge, affecting over 10 million individuals, with prevalence expected to double by 2040. Early diagnosis remains difficult due to the late emergence of motor symptoms and limitations of traditional clinical assessments. In this study, we propose a novel pipeline that leverages keystroke dynamics as a non-invasive and scalable biomarker for remote PD screening and telemonitoring. Our methodology involves three main stages: (i) preprocessing of data from four distinct datasets, extracting four temporal signals and addressing class imbalance through the comparison of three methods; (ii) pre-training eight state-of-the-art deep-learning architectures on the two largest datasets, optimizing temporal windowing, stride, and other hyperparameters; (iii) fine-tuning on an intermediate-sized dataset and performing external validation on a fourth, independent cohort. Our results demonstrate that hybrid convolutional-recurrent and transformer-based models achieve strong external validation performance, with AUC-ROC scores exceeding 90% and F1-Score over 70%. Notably, a temporal convolutional model attains an AUC-ROC of 91.14% in external validation, outperforming existing methods that rely solely on internal validation. These findings underscore the potential of keystroke dynamics as a reliable digital biomarker for PD, offering a promising avenue for early detection and continuous monitoring.
Paper Structure (10 sections, 1 figure, 5 tables)

This paper contains 10 sections, 1 figure, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Method and Experimental Setup
  4. Datasets and Pre-processing
  5. Pre-training
  6. Fine-tuning and External Validation
  7. Results and Discussions
  8. Conclusion
  9. Acknowledgments.
  10. Disclosure of Interests.

Figures (1)

  • Figure 1: Schematic representation of the proposed method, comprising four steps: dataset pre-processing, pre-training, fine-tuning, and external validation. In the NN (Neural Network) and FC (Fully Connected) blocks, weight freezing and updating are represented by ice and flame icons, respectively. Class balancing is illustrated with orange for the majority class (“Maj.”) and light blue for the minority class (“Min.”). In the IMBALMED francesconi2025class strategy, multiple sub-datasets with different class distributions are created.