New machine learning program accurately predicts who will stick with their exercise program

Staying active is one of the most important things you can do for your health. Regular exercise helps you live longer, lowers your risk of disease, improves your mood, and boosts energy levels. But only a small portion of people actually meet exercise recommendations. So, what makes someone stay committed to working out?

A team of researchers set out to find the answer. At the University of Mississippi, scientists analyzed national health data using machine learning to find patterns in who meets physical activity (PA) guidelines and why. This approach could help doctors and trainers better support your health by understanding what motivates people like you to keep moving.

This study, published in the journal, Scientific Reports, looked at data collected between 2009 and 2018 from the National Health and Nutrition Examination Survey, a large U.S. survey that tracks health and diet habits. The research team included doctoral students Seungbak Lee and Ju-Pil Choe, and Professor Minsoo Kang. They used a tool called machine learning to sort through over 30,000 survey responses.

Machine learning helps computers find patterns in large amounts of data. Unlike older statistical tools, which expect clean, linear data, machine learning works well even when the data is messy or complicated. It can sort out which pieces of information matter most in predicting behavior, like who sticks with exercise routines.

The researchers filtered the data to include only people age 18 and older without diseases that could limit exercise, such as cancer, diabetes, or arthritis. After removing entries with missing answers, the final data set included 11,638 participants.

Each person’s responses were grouped into three main areas: demographics (age, gender, race, income, etc.), body measurements (like body mass index and waist size), and lifestyle habits (such as alcohol use, smoking, sleep, and sedentary time). The goal was to build models that could predict whether someone met the weekly activity guidelines.

According to U.S. health officials, adults should get at least 150 minutes of moderate exercise or 75 minutes of intense activity each week. Unfortunately, the average American only gets about two hours of activity weekly—half of what’s recommended.

Using six different machine learning algorithms, the researchers built 18 prediction models to test various combinations of factors. These models were measured by how accurate they were, how well they could find patterns, and how balanced their predictions were.

The best-performing model was a decision tree using all available variables. It had an accuracy of about 70.5% and an F1 score (a balance between precision and recall) of 0.819. That means it correctly predicted who met exercise guidelines most of the time.

But beyond just performance, the team wanted to know which specific factors were most useful in making predictions. Using a technique called Permutation Feature Importance (PFI), they found that sedentary behavior, age, gender, and educational status were the most important predictors. Even though some models gave slightly different answers, these factors kept showing up again and again.

Ju-Pil Choe explained, "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. While factors like gender, BMI and age are more innate to the body, educational status is an external factor."

The team noted that people who sat for long periods, had lower education levels, or were of a certain gender were less likely to meet activity guidelines. This helps explain who is more likely to stick with physical activity and why. These insights could guide future programs aimed at helping people develop healthier habits.

While the results are promising, the researchers did note some limits to their approach. One key issue is that the survey data relied on self-reported activity levels. People often overestimate how much they exercise when asked to recall it from memory.

"One limitation of our study was using subjectively measured physical activity data," Choe said. "More accurate, objective data would improve the study's reliability."

Future research could fix this by using wearable fitness trackers or apps that automatically log physical activity. Machine learning could then use that objective data to find even stronger and more detailed patterns.

Despite this limitation, the research shows that machine learning has great promise for studying health behaviors. It doesn’t just tell us what the trends are—it helps uncover why those trends exist in the first place.

Why is all this important? Because understanding the reasons behind someone’s exercise habits can help health professionals create better, more personalized plans. Instead of giving the same advice to everyone, doctors could use data-driven models to figure out what motivates each person.

For example, if someone has a sedentary job and low education levels, they might need more support or different types of motivation to stay active. Knowing that these factors matter allows experts to build programs that work for each individual.

This is especially helpful for trainers, coaches, and even health app developers. They can create exercise routines that feel more achievable and are tailored to your lifestyle, age, and daily habits. It makes sticking with a workout plan easier and more realistic.

Professor Kang summed up the purpose of the study: "Physical activity adherence to the guidelines is a public health concern because of its relationship to disease prevention and overall health patterns. We wanted to use advanced data analytic techniques, like machine learning, to predict this behavior."

Other studies have already used machine learning in related ways. For example, some researchers built models to classify physical activity in children using motion sensors. Others used neural networks to sort activity levels based on body movements. But this study is one of the first to focus on predicting adherence to activity guidelines using only self-reported data and a wide mix of demographic, body, and lifestyle factors.

The results show that machine learning can be a powerful tool in public health. It reveals patterns that might be invisible with traditional methods. And it gives researchers a new way to help people live healthier, longer lives.

Note: The article above provided above by The Brighter Side of News.

Like these kind of feel good stories? Get The Brighter Side of News' newsletter.