What researchers studied
An Indian research team designed a system for personalised exercise and fitness recommendations.
It combined supervised machine-learning models with a separate rule-based reasoning component.
Inputs included demographic information, body composition, cardiovascular measures, exercise habits and previous experience.
Prediction models estimated calorie expenditure and classified exercise activity levels.
The rules layer translated those outputs into suggested exercise intensity and fitness objectives.
Researchers compared regression and classification methods, including tree-based approaches, and examined the importance of input features.
This was a technical modelling evaluation rather than a clinical intervention or trial of long-term fitness outcomes.
What they found
The researchers reported promising performance in predicting energy expenditure and classifying activity levels.
Their comparisons included random forest and XGBoost regression models alongside classification approaches.
The system also generated example personalised recommendations from predicted values and explicit rules.
The combined design may make certain decisions more interpretable than recommendations produced entirely by opaque statistical models.
The paper does not establish that users became fitter, maintained exercise habits longer or suffered fewer injuries.
Prediction performance and sample recommendations therefore cannot be treated as evidence of improved health.
The technical evaluation is a preliminary step towards possible application development.
What it means
A hybrid design could be useful when a fitness app needs both statistical estimates and understandable decision rules.
Explicit rules may help developers inspect why a recommendation was generated.
Nevertheless, exercise advice must account for medical conditions, physical limitations and changing symptoms that may not be represented in model inputs.
The study supports additional evaluation rather than replacing professional clinical or exercise guidance.
Important limitation
The evaluation did not establish prospective health benefits or real-world safety across diverse users.
Accuracy can change when models encounter new populations, sensors or measurement conditions.
The example exercise outputs are not clinically validated prescriptions and should not be described as such.
Source Information
Study: AI-based personalized exercise and health recommendation system using hybrid modeling
Authors: Sarthak Singh, Abhijeet Singh, Ayush Tiwari, Himanshu Gurjar, Siddhi Patel and Amanpreet Kaur
Journal / Source: Frontiers in Artificial Intelligence
Published: 12 November 2026







