INTERPRETABLE MACHINE LEARNING ANALYSIS OF DIET, STRESS, AND THEIR IMPACT ON DIABETES RISK: IMPLICATIONS FOR METABOLIC HEALTH AND PHYSICAL PERFORMANCE

Abstract

Background: Diabetes is a major metabolic disorder that not only affects overall health but also significantly influences physical activity levels, athletic performance, and exercise recovery. Early and accurate detection of diabetes is crucial for preventing complications, optimizing metabolic function, and enhancing participation in physical activity and sports. This study explores the impact of dietary habits and stress factors on diabetes risk using interpretable machine learning models, with a focus on their implications for sports science, rehabilitation, and metabolic health management. Methods: Machine learning models were developed using dietary intake data combined with positive and negative stress indicators to enhance predictive accuracy for diabetes detection. Comparative analyses were conducted to evaluate the relative impact of diet and stress on diabetes risk, with an emphasis on metabolic efficiency, energy regulation, and physical endurance. Random Forest and other interpretable machine learning approaches were applied to ensure transparency in the prediction process, enabling clinicians, sports scientists, and health practitioners to derive actionable insights from the results. Results: The inclusion of stress-related features significantly improved model accuracy and generalizability, highlighting the interplay between psychological stress, metabolic function, and physical performance. Contrary to traditional assumptions, positive stress exhibited a stronger influence on diabetes risk than negative stress, suggesting that psychological resilience and adaptive stress responses play a crucial role in metabolic adaptation and physical health. Additionally, dietary factors, particularly carbohydrate intake, emerged as the most critical determinant of diabetes risk, reinforcing the importance of nutritional regulation in sports performance and metabolic optimization. The proposed machine learning model achieved an accuracy exceeding 99%, demonstrating its potential as a reliable tool for early diabetes detection, personalized intervention, and sports health management. Conclusions: This study provides valuable insights into the role of diet and stress in diabetes risk and their implications for physical activity, sports participation, and athletic performance. The findings highlight the need for integrated lifestyle interventions that combine nutritional optimization, stress management, and structured exercise programs to enhance metabolic resilience and athletic endurance. By leveraging interpretable machine learning, healthcare professionals and sports scientists can develop personalized strategies for diabetes prevention, physical conditioning, and performance enhancement. Future research should explore the long-term impact of diet-stress interactions on sports performance and recovery in diabetic and prediabetic populations.