Prediction of Chronic Disease Using Explainable Hybrid Machine Learning

In this study, a 28-item questionnaire developed by the researchers (Aysun Atacan and Gülşen Taşkın) will be used as the data collection tool. Since the intellectual property rights and developer identity belong to the researchers, there are no copyright or usage restrictions. Containing both qualitative and quantitative evaluation questions, the questionnaire consists of 5 main sections: personal information, daily life assessment, walking level assessment, household chores assessment, and work/workplace assessment. Eight of the questions cover sociodemographic information, while 20 focus on determining and evaluating physical activity levels in daily life, walking, home, work, and transportation. Participants are asked to answer the questions by considering their lifestyle over the past month. Based on these data-recorded according to how many days per week and how many minutes per day the activities are performed-calculations will be made by multiplying the MET value, frequency (days/week), and duration (minutes/day) to obtain the "MET-min/week" score. Participants will be categorized into three groups based on their total weekly MET expenditures: <600 MET-min/week as inactive, 600-3000 MET-min/week as minimally active, and >3000 MET-min/week as health-enhancing physically active (sufficiently active); and into groups based on their average daily step counts: <5000 as sedentary/inactive, 5001-7499 as low active, 7500-9999 as somewhat active, 10000-12499 as active, and >12500 as highly active. Chronic disease risk predictions will then be evaluated via machine learning based on these physical activity levels and step counts.

The primary aim of this research is to quantitatively demonstrate the impact of physical activity levels and lifestyle habits on chronic disease risk factors using machine learning (ML) methods. The models to be developed aim not only for high predictive performance but also for the clinical interpretation of which physical activity parameters are more decisive on disease risk, utilizing explainable artificial intelligence methods such as SHapley Additive exPlanations (SHAP). Consequently, the goal is to establish a decision support mechanism for the early detection of individuals at risk and to place personalized exercise prescriptions on a scientific foundation.