Parkinson’s Disease Detection Using Keystroke Dynamics and with Convolutional Neural Networks

Resumen

Parkinson’s Disease is a progressive neurological disorder. It poses major challenges for early diagnosis because its initial symptoms are subtle and depend on subjective clinical evaluations. However, an early diagnosis becomes even more challenging because traditional methods have relied so much on costly medical tests and subjective clinical assessments that often fail to detect PD at its earliest stages when symptoms are minimal. In this respect, this study develops a deep learning-based framework for the automatic detection of Parkinson’s Disease using keystroke dynam- ics collected from typing interactions. Keystroke timing features such as hold time and latency are excellent digital biomarkers for early PD iden- tification since they carry fine-grained motor-control information. The paper proposes a deep learning system using Convolutional Neural Net- works (CNNs) model that identifies users as PD or Non-PD based on real- world typing behaviours. The effectiveness of the suggested Pure 1D CNN model in identifying discriminative temporal patterns from keystroke data for Parkinson’s disease detection is demonstrated by its 95.21% accuracy and 99% ACC value. Besides the deep learning model, machine learning algorithms were also implemented in order to compare the performance and confirm the robustness of the proposed system. This system aims to improve diagnostic accuracy by reducing dependency on expensive clinical procedures and enabling continuous remote monitoring.