Detection of Parkinson's disease with keystroke data

Area of Science:

• Neurology
• Computer Science
• Biomedical Engineering

Background:

• Parkinson's disease (PD) is a prevalent neurological disorder characterized by nerve damage and motor impairments like tremors and bradykinesia.
• Current diagnostic methods can be invasive or lack early detection capabilities, highlighting the need for novel approaches.

Purpose of the Study:

• To develop and evaluate machine learning (ML)-based models for Parkinson's disease detection using keyboard keystroke dynamics.
• To identify key features from raw keystroke data that effectively distinguish individuals with PD.

Main Methods:

• Extracted 378 features from raw keystroke data across 14 sub-datasets, categorized by drug use, disease severity, and gender.
• Developed and compared ML models using Support Vector Machines (SVM), k-Nearest Neighbors (kNN), and Random Forest (RF) algorithms.
• Employed feature selection techniques including Minimum Redundancy Maximum Relevance (mRmR), RELIEF, and sequential forward selection (SFS), alongside Random Forest (RF).
• Created ensemble models: Feature Ensemble (FE) combining popular features and Model Ensemble (ME) combining multiple ML algorithms via majority vote.

Main Results:

• Feature ensemble (FE) models achieved high accuracy, ranging from 91.73% to 100% across the 14 datasets.
• Model ensemble (ME) models demonstrated strong performance, with accuracies ranging from 81.08% to 100%.

Conclusions:

• Keystroke dynamics combined with machine learning offer a promising, non-invasive method for early Parkinson's disease detection.
• Ensemble approaches, particularly the feature ensemble, show significant potential for accurate PD diagnosis based on typing patterns.