Welcome to Movement Health Predictor

Advanced Deep Learning System for Movement-Related Health Risk Prediction

Recent advances in wearable technology enable noninvasive, low-cost characterization of sleep, sedentary behavior, light physical activity, and moderate-to-vigorous physical activity (MVPA) at high temporal resolution and population scale.

To fully harness the potential of these data, we propose a comprehensive Behavioral Clock framework and introduce the Physical Pan-Disease Model (PPDM)—a high-resolution deep learning architecture that integrates long short-term memory (LSTM) networks with multi-head self-attention to model accelerometer-derived activity sequences from nearly 100,000 UK Biobank participants.

By capturing hourly behavioral patterns and explicitly accounting for weekday–weekend structure, PPDM reflects realistic variation in daily human activity. These temporally resolved behavioral phenotypes are linked to 370 incident diseases, resulting in an interpretable pan-disease behavioral atlas that enables systematic assessment and stratification of disease susceptibility based on behavioral rhythms.

Model generalizability is demonstrated through external validation in the NHANES cohort, where PPDM robustly stratifies mortality risk. To provide biological grounding, we integrate Mendelian randomization analyses with inflammatory, metabolic, and proteomic profiles, yielding convergent evidence for behavioral–disease associations and their underlying molecular pathways.

Finally, we develop a counterfactual inference framework to identify population-level and disease-specific behavioral modifications with potential for risk reduction, and release an open-access interactive resource for exploration of behavioral profiles and model-based risk reports.

Movement Health Prediction System Architecture
370
Diseases Covered
4*24*2
Activity Analysis for 4 Sports Mode in Two Days

Comprehensive Health Risk Assessment Platform

The system processes 192 time-series features representing hourly activity distributions across weekdays and weekends, categorized into four distinct activity types: sedentary behavior, light activity, moderate-to-vigorous activity, and sleep patterns. This granular analysis enables precise investigation of individual activity patterns and personalized health risk assessment.

Dual Model Framework: Our platform offers two sophisticated prediction frameworks:

External validation on an independent cohort of 800 participants demonstrated exceptional real-world applicability. The model architecture employs advanced deep learning techniques, combining Long Short-Term Memory (LSTM) networks for temporal pattern recognition with Self-Attention mechanisms for identifying critical time periods and activity patterns.

Feature importance analysis revealed that patterns including prolonged sedentary bouts, irregular sleep schedules, and insufficient moderate-to-vigorous physical activity were key predictors of health risks. The system specifically identifies risk factors for metabolic disorders, cardiovascular issues, and musculoskeletal problems across 370 ICD-10 coded diseases.

To use four movement patterns to description the person's behavior, Collecting data on the time spent in four behavior patterns over 24 hours for two days, and use it to predict a person's risk of illness.

The implementation of this prediction system significantly improves early detection of activity-related health risks, enables personalized intervention strategies, and enhances population health through targeted activity recommendations. The model's ability to identify at-risk individuals based on daily movement patterns provides valuable insights for preventive healthcare.

Start Your Health Assessment View Professional Analysis