KeyGAN: Synthetic keystroke data generation in the context of digital phenotyping.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Digital Health

Background:

• Developing reliable digital biomarkers for psychomotor impairment due to neurodegeneration is crucial.
• Existing methods for generating synthetic keystroke data may lack the realism needed for complex applications.
• Fine motor control and cognitive processes influence typing kinematics, offering potential insights into neurological conditions.

Purpose of the Study:

• Introduce and evaluate KeyGAN, a novel generative modeling-based synthesizer for realistic keystroke data.
• Assess KeyGAN's ability to capture the nuances of natural typing for biomarker development.
• Demonstrate KeyGAN's utility in applications such as biometric authentication and characterizing Parkinson's Disease.

Main Methods:

• KeyGAN was designed to ensure high realism in synthetic keystroke feature distributions.
• Evaluated KeyGAN's synthetic data quality using TypeNet (biometric authentication) and nQiMechPD (Parkinson's Disease characterization) as referee controls.
• Compared KeyGAN's performance against a reference random Gaussian generator.

Main Results:

• KeyGAN successfully fooled the TypeNet biometric authentication system, outperforming the reference comparator.
• KeyGAN demonstrated superior approximation to real typing data in characterizing Parkinson's Disease using nQiMechPD.
• Synthetic data from KeyGAN could replace nearly 20% of real Parkinson's Disease samples in training sets without performance loss.
• Low Fréchet Distance (<0.03) and Kullback-Leibler Divergence (<700) confirmed the high fidelity of KeyGAN-generated data.

Conclusions:

• KeyGAN shows significant potential as a realistic keystroke data synthesizer for neurological disorder research.
• The model effectively reproduces complex typing patterns relevant to digital biomarkers.
• KeyGAN's synthetic data can supplement real data for training algorithms, advancing digital biomarker research for neurodegenerative and psychomotor disorders.