Exploring machine learning approaches for early diabetes risk prediction: A comprehensive examination of health indicators and models

Abstract

Background: The increased prevalence of morbidity and mortality associated with Type 2 diabetes is due to changing lifestyles, demanding improved disease management measures. To tackle this, scientists are increasingly looking to technological advances, notably machine learning, for illness prevention and management, particularly in non-communicable diseases. The emphasis is on establishing an early detection system to identify Type 2 diabetes risk factors, enabling prompt treatments and preventative steps to reduce the disease’s rising prevalence. Materials and methods: The research aimed to assess the association of diabetes class with health indicators. Five machine learning models were employed with cross-validation techniques to predict early diabetes risk. The performance matrices of the models were evaluated and compared with the existing work. Results: In multivariate analysis, we found polyuria (β=3.492; Aor=32.872; 95% CI=11.09,97.35; p<0.001), polydipsia (β=-4.100; Aor=60.378; 95%CI=18.28,199.37; p<0.001), polyphagia (β=1.181; Aor=3.25; 95%CI=1.23,8.57; p=0.017), genital thrush (β=1.08; Aor=2.96; 95%CI=1.26,7.53; p=0.023), irritability (β=2.28; Aor=9.82; 95%CI=3.41,28.26; p<0.001), and partial paresis (β=1.2406; Aor=3.45; 95% CI=1.35,8.79; p=0.009) are the potential health risk indicators for positive diabetes class. Conclusion: Using an interpretable feature learning approach for early diabetes prediction improves the use of global health data. This method forecasts hazards correctly and gives insights into influential aspects. As a result, a more proactive healthcare strategy is implemented, allowing for more prompt treatments and encouraging a more hopeful future by improving patient outcomes and lowering the total burden of diabetes on individuals and healthcare systems.