Summary: A new study has used machine learning to identify the key predictors of physical activity adherence, analyzing data from nearly 12,000 individuals. The research found that time spent sitting, gender, and education level were the strongest indicators of whether someone met weekly exercise guidelines.
By training models on lifestyle, demographic, and health survey data, researchers could predict exercise habits more flexibly than traditional approaches. These insights could inform more effective fitness recommendations and public health strategies tailored to individual needs.
Key Facts:
- Top Predictors: Sedentary time, gender, and education level were the most consistent predictors of exercise adherence.
- Study Scope: Researchers used machine learning on data from 11,683 participants in a national health survey.
- Potential Impact: Findings could improve personalized workout plans and inform health policy.
Source: University of Mississippi
Sticking to an exercise routine is a challenge many people face. But a University of Mississippi research team is using machine learning to uncover what keeps individuals committed to their workouts.
The team – Seungbak Lee and Ju-Pil Choe, both doctoral students in physical education, and Minsoo Kang, professor of sport analytics in the Department of Health, Exercise Science and Recreation Management – hopes to predict whether a person is meeting physical activity guidelines based on their body measurements, demographics and lifestyle.
They have examined data from about 30,000 surveys. To quickly sort through such a huge data set, they’ve turned to machine learning, a way of using computers to identify patterns and make predictions based on the information.
The group’s results, published in the Nature Portfolio journal Scientific Reports are timely, Kang said
“Physical activity adherence to the guidelines is a public health concern because of its relationship to disease prevention and overall health patterns,” he said.
“We wanted to use advanced data analytic techniques, like machine learning, to predict this behavior.”
The Office of Disease Prevention and Health Promotion, part of the U.S. Department of Health and Human Services, suggests that adults should aim for at least 150 minutes of moderate exercise, or 75 minutes of vigorous exercise, each week as part of a healthy lifestyle.
Research shows that the average American spends just two hours per week on physical activity – half of the four hours recommended by the Centers for Disease Control and Prevention.
Lee, Choe and Kang used public data from the National Health and Nutrition Examination Survey, a government-sponsored survey, covering 2009-18.
“We aimed to use machine learning to predict whether people follow physical activity guidelines based on questionnaire data, and find the best combination of variables for accurate predictions,” said Choe, the study’s lead author.
“Demographic variables such as gender, age, race, educational status, marital status and income, along with anthropometric measures like BMI and waist circumference, were considered.”
The researchers also considered lifestyle factors including alcohol consumption, smoking, employment, sleep patterns and sedentary behavior to understand their impact on a person’s physical activity, he said.
The results showed that three key factors – how much time someone spends sitting, their gender, and their education level – showed up consistently in all the top-performing models that predict exercise habits, even though each model identified different variables as important.
According to Choe, these factors are especially important for understanding who is more likely to stay active and socially connected, and they could help guide future health recommendations.
“I expected that factors like gender, BMI, race or age would be important for our prediction model, but I was surprised by how significant educational status was,” he said. “While factors like gender, BMI and age are more innate to the body, educational status is an external factor.”
During the analysis, the researchers excluded data from people with certain diseases and responses missing physical activity data. That culled the relevant data to 11,683 participants.
The researchers say machine learning gives them more freedom to study the data. Older methods expect things to follow a straight-line pattern, and they don’t work well when some pieces of information are too similar.
Machine learning doesn’t have those limits, so it can find patterns with greater flexibility.
“One limitation of our study was using subjectively measured physical activity data, where participants recalled their activity from memory,” Choe said.
“People tend to overestimate their physical activity when using questionnaires, so more accurate, objective data would improve the study’s reliability.”
Because of this, the researchers say they could use a similar method for future research in this area, but explore different factors, including dietary supplements use, using more machine learning algorithms or relying on objective data instead of self-reported information.
That could help trainers and fitness consultants produce workout regimens that people can actually stick with for the long haul.
About this AI and exercise research news
Author: Clara Turnage
Source: University of Mississippi
Contact: Clara Turnage – University of Mississippi
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Machine learning modeling for predicting adherence to physical activity guideline” by Seungbak Lee et al. Scientific Reports
Abstract
Machine learning modeling for predicting adherence to physical activity guideline
This study aims to create predictive models for PA guidelines by using ML and examine the critical determinants influencing adherence to the PA guidelines. 11,638 entries from the National Health and Nutrition Examination Survey were analyzed.
Variables were categorized into demographic, anthropometric, and lifestyle categories. 18 prediction models were created by 6 ML algorithms and evaluated via accuracy, F1 score, and area under the curve (AUC).
Additionally, we employed permutation feature importance (PFI) to assess the variable significance in each model.
The decision tree using all variables emerged as the most effective method in the prediction for PA guidelines (accuracy = 0.705, F1 score = 0.819, and AUC = 0.542).
Based on the PFI, sedentary behavior, age, gender, and educational status were the most important variables.
These results highlight the possibilities of using data-driven methods with ML in PA research.
Our analysis also identified crucial variables, providing valuable insights for targeted interventions aimed at enhancing individuals’ adherence to PA guidelines.
