A Machine Learning Approach for Prediction of Sedentary Behavior Based on Daily Step Counts

Area of Science:

• Health Informatics
• Machine Learning
• Public Health

Background:

• Sedentary behavior is a significant public health concern, associated with chronic diseases and increased mortality.
• Accurate prediction of sedentary behavior is essential for developing effective interventions.

Purpose of the Study:

• To propose and evaluate a machine learning approach for predicting next-day sedentary behavior using step count data.
• To compare the performance of Logistic Regression, Random Forest, XGBoost, Convolutional Neural Networks, and an ensemble model.

Main Methods:

• Utilized a crowd-sourced dataset of step count data from 33 users over one month for training and testing.
• Employed an additional dataset from one user over six months for further evaluation.
• Developed and assessed an ensemble model combining multiple machine learning algorithms.

Main Results:

• All tested models successfully predicted next-day sedentary behavior.
• The Majority Vote Ensemble model demonstrated superior performance, outperforming individual algorithms.
• The ensemble model achieved 82.12% accuracy on a multi-subject dataset and 76.88% on an unseen dataset, with improved false positive reduction.

Conclusions:

• Machine learning, particularly ensemble methods based on step count data, shows promise for accurate sedentary behavior prediction.
• The developed approach can inform the creation of intelligent systems to combat sedentary lifestyles.
• Further development of AI-driven tools can aid in public health initiatives targeting sedentary behavior.