# Unveiling the Digital Phenotype of Physical Activity Behavior in Community-Dwelling Older Adults Using Machine Learning

## TL;DR

This study uses machine learning to predict physical activity patterns in older adults, combining wearable data with psychological factors.

## Contribution

The study introduces a novel combination of wearable sensor data and machine learning to predict and understand physical activity in older adults.

## Key findings

- Psychological factors like self-efficacy are key predictors of physical activity levels in cross-sectional analysis.
- Longitudinal models using a week of step count data reliably forecast future physical activity.
- Machine learning and deep learning methods provide valuable insights into physical activity behaviors in older adults.

## Abstract

Physical activity (PA) is an important factor for maintaining health and well-being, especially in older adults. This study aims to apply machine learning methods to predict PA patterns and identify key factors influencing these behaviors among community-dwelling older adults. Linear and Logistic Regression, Elastic Net, and Light Gradient Boosting Machine (LightGBM) models were used to analyze cross-sectional data. While longitudinal data collected over 14 days were analyzed using LightGBM, Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM). The most important predictors identified in the cross-sectional analysis were the Exercise Self-efficacy Scale (ESES) for PA levels and the Geriatric Depression Scale (GDS) for the International Physical Activity Questionnaire (IPAQ) as a continuous measurement. In the longitudinal analysis, using a seven-day sequence of step count data provided the best performance for forecasting physical activity for the entire next day. Overall, the findings indicate that combining wearable sensor data with machine learning and deep learning methods can provide valuable insights into physical activity behaviors among older adults. In the cross-sectional analysis, psychological and motivational factors such as self-efficacy were identified as important factors for activity levels, while in the longitudinal analysis, using a week of past step count data provided the most reliable predictions of future-day physical activity.