Mathematical Foundations of Diabetes Forecasting Studies: A Comparative Analysis of Statistics and ML Models

This study provides a systematic comparison of the mathematical properties, strengths, and limitations of traditional statistical methods and machine learning models in diabetes forecasting. While classical approaches like logistic regression and ANOVA offer interpretability and simplicity, their reliance on linear assumptions and sensitivity to heteroscedasticity limit their utility in modeling complex, nonlinear relationships inherent in diabetes data. In contrast, machine learning techniques-including neural networks, random forests, and gradient boosting-excel in capturing high-dimensional interactions and nonlinear dynamics, achieving superior predictive accuracy. However, these gains come at the cost of computational complexity, black box interpretability challenges, and ethical concerns around algorithmic bias. Through a detailed analysis of mathematical frameworks (e.g., activation functions, regularization, ensemble methods), we demonstrate how hybrid approaches integrating explainable AI (XAI) can bridge the gap between statistical rigor and clinical usability. Our findings highlight the critical trade-offs between model interpretability, predictive power, and scalability, offering actionable insights for optimizing diabetes risk prediction in precision medicine.

کلیدواژه ها:

Statistical modeling ، Machine learning ، Predictive analytics ، Sample size ، Big data