Walking fingerprinting.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Individual identity can be predicted from accelerometry data during walking.
• Previous work transformed time-series data into images for prediction.
• Predictors were derived from image grid cells using logistic regression.

Purpose of the Study:

• Implement machine learning for identity prediction from accelerometry.
• Develop inferential methods to identify predictive features.
• Create a multivariate functional regression model to improve prediction accuracy.

Main Methods:

• Machine learning algorithms applied to grid cell predictors.
• Statistical methods for screening predictive grid cells.
• Multivariate functional regression model developed to avoid predictor space partitioning.

Main Results:

• High accuracy (≥95% rank-1) achieved in a 32-individual study.
• Variable accuracy (41%-98%) observed in a 153-participant study based on method.
• Insights gained into factors influencing individual predictability.

Conclusions:

• Machine learning and novel regression models enhance identity prediction from walking accelerometry.
• The developed methods offer improved accuracy and feature selection capabilities.
• Findings contribute to understanding individual variability in movement data.