Related Experiment Video
Updated: Jun 23, 2025
10:28
Dynamic Digital Biomarkers of Motor and Cognitive Function in Parkinson's Disease
Published on: July 24, 2019
15.2K
Framework for Ranking Machine Learning Predictions of Limited, Multimodal, and Longitudinal Behavioral Passive
Tahsin Mullick1, Sam Shaaban2, Ana Radovic3
1Department of Systems and Information Engineering, University of Virginia, Charlottesville, VA, United States.
JMIR AI
|June 14, 2024
Summary
This study introduces a new framework for analyzing small, noisy sensor data from mobile health studies. The FLMS model improves prediction accuracy by combining user-agnostic and personalized approaches for better health monitoring.
Area of Science:
- Digital Health
- Mobile Health (mHealth)
- Biomedical Data Science
Background:
- Passive mobile sensing offers remote health monitoring but faces challenges with small, noisy, and incomplete longitudinal data.
- Current machine learning (ML) models struggle with limited data, often requiring a trade-off between user-agnostic and personalized approaches, leading to suboptimal predictions.
Purpose of the Study:
- To develop a novel framework (FLMS) for processing and predicting outcomes from small, multimodal, longitudinal sensor data in health studies.
- To integrate user-agnostic and personalized modeling strategies with advanced ranking techniques to enhance prediction accuracy.
Main Methods:
- Introduced the Framework for Longitudinal Multimodal Sensors (FLMS) incorporating tensor-based aggregation and ranking.
- Implemented sensor fusion techniques and balanced user-agnostic with personalized modeling using cross-validation.
- Validated the FLMS framework on a real-world dataset of adolescents with major depressive disorder.
Main Results:
- The FLMS framework demonstrated a 7% increase in accuracy and a 13% increase in recall on the real-world dataset.
- Compared to state-of-the-art ML algorithms, FLMS showed an 11% accuracy improvement for depression prediction, effectively handling data sparsity and overfitting.
Conclusions:
- The FLMS framework addresses the critical gap in modeling small passive sensor datasets for health applications.
- By synergistically combining user-agnostic and personalized modeling with effective prediction filtering, FLMS enhances the utility of mobile sensing in clinical research.
