Fitness trackers and smartwatches promise to help people understand their health, but accurately measuring physical activity remains a persistent challenge. Many wearable devices use proprietary algorithms, making it difficult for researchers to evaluate how exercise and energy expenditure are actually calculated and how reliable those measurements are.
This summer, Dayyan Chaudhri ’28, a computer science major in Loyola Marymount University’s Frank R. Seaver College of Science and Engineering, explored how machine learning and wearable sensors can help make health monitoring more accurate, transparent, and accessible.
Working alongside Delaram Yazdansepas, assistant professor of computer science, Chaudhri participated in Seaver College’s Summer Opportunities for Advanced Research (SOAR) program, where undergraduate students collaborate with faculty mentors on original research projects.
At the center of the project was a simple question: Can a sensor that already exists in nearly every smartphone provide trustworthy information about physical activity?
“Right now, getting activity right takes money — lab equipment, or a chest strap, or a watch that costs a few hundred dollars,” Chaudhri said. “Almost everyone owns an accelerometer, because it’s already in every phone, it barely uses battery, and it runs all day without being asked.”
Using data collected from wearable accelerometers, Chaudhri compared two widely accepted approaches for estimating physical activity intensity. One method converts the amount of wrist movement detected by a sensor into an estimate of exertion, while the other identifies a person’s activity, such as walking, jogging, or climbing stairs, and assigns an effort level based on that activity.
Because both approaches are commonly used throughout health and fitness research, Chaudhri wanted to understand whether they produced similar results when applied to the same dataset. The answer was surprising.
After analyzing data from 33 participants, he found that the two methods disagreed roughly one-third of the time when determining whether someone had reached moderate-to-vigorous levels of physical activity.
“Put plainly, how much exercise a watch credits you with can shift by about a third on the method alone,” Chaudhri said.
The findings are helping inform the development of In2Eat, an intuitive eating app created by Yazdansepas, Mandy Korpusik, assistant professor of computer science, and Hawley Almstedt, professor of health and human sciences. Chaudhri explained that In2Eat will report activity as recognized behavior, such as walking, stairs, and jogging, rather than a single moderate-to-vigorous physical activity total that can shift depending on the measurement method.
During the summer, Chaudhri led the integration phase of the project. He developed a system that processed the same accelerometer data through both estimation methods and compared the results using a standard benchmark for moderate physical activity. He explained, “I wrote the pipeline that scores the same wrist signal through both routes and compares them at the 3 MET (metabolic equivalent of task) line.”
Beyond the technical training, the summer experience transformed how he thinks about research.
“I came in assuming research meant more data and better tests,” Chaudhri said. “But the most important thing I found all summer was already sitting in the arithmetic, waiting for someone to look.”
That realization came after discovering a structural flaw in one of the accepted methods being evaluated. The experience reinforced a lesson that has stayed with him throughout the research process.
“The lesson: Ask what a method is capable of saying before you trust what it said,” he said.
One of the most rewarding aspects of the summer for Chaudhri was presenting his work to different audiences. At Seaver College’s SOAR 10-Year Anniversary Celebration, Chaudhri presented a project poster to the larger LMU community. “I presented the work to an audience outside my direct discipline, which required me to articulate the project without relying on technical vocabulary,” he added.
Later, in July, he presented the project at the Computer Science, Computer Engineering, and Applied Computing (CSCE) Conference in Las Vegas to a room full of experts. “At CSCE, I delivered a 15-minute talk to a room full of researchers working in this field, who gave me new ideas to level up the next phases of this research,” Chaudhri said. “Many were amazed at the work we have done thus far and provided invaluable recommendations and questions.”
As an incoming junior and member of the class of 2028, Chaudhri had the opportunity not only to conduct faculty-mentored research through Seaver College’s summer research program but also to present his findings at a national professional conference, an experience typically associated with more advanced researchers.
“I came in curious and driven and left with new ideas of what’s next,” he said.