Digital Phenotyping of Sensation Seeking: A Machine Learning Approach Using Gait Analysis

Area of Science:

• Behavioral science
• Computational neuroscience
• Digital health

Background:

• Sensation seeking is a risk factor for mental health disorders and maladaptive behaviors.
• Traditional self-report measures for sensation seeking have limitations.
• Objective assessment methods are needed to quantify sensation-seeking traits.

Purpose of the Study:

• To introduce and validate a digital phenotyping approach for quantifying sensation-seeking traits.
• To combine computational gait analysis with machine learning (ML) for objective assessment.
• To explore the potential of movement dynamics in understanding psychological indicators.

Main Methods:

• Collected natural gait sequences and self-report data (Brief Sensation-Seeking Scale for Chinese, BSSS-C) from 233 healthy adults.
• Utilized computer vision (OpenPose) to extract skeletal keypoints, processed into kinematic data.
• Developed and compared three ML models (SMO Regression, Multilayer Perceptron, Bagging) using 300 temporospatial gait features and 10-fold cross-validation.

Main Results:

• The SMO Regression model achieved the best performance with a correlation coefficient (r) of 0.60, MAE of 3.50, RMSE of 4.59, and R² of 0.26.
• The gait-based digital phenotyping approach demonstrated proof-of-concept for assessing sensation seeking.
• The study successfully transformed basic movement patterns into meaningful psychological indicators.

Conclusions:

• Gait-based digital phenotyping offers a scalable and objective method for assessing sensation seeking.
• This approach has potential applications in clinical screening and behavioral research.
• The study advances behavioral biometrics by linking movement dynamics to psychological traits.