Dual Approach to Type 2 diabetes Mellitus Risk Assessment in Women: Machine Learning Predictions and Fractional-Order Modeling of Physiological Dynamics

This study examines diabetes risk in women through predictive machine learning models and fractional- order physiological modeling. Machine learning models, including Bagged Trees, k-Nearest Neighbors (k-NN), Decision Trees, SVM (Support Vector Machine), and Logistic Regression, were applied to assess accuracy in diabetes prediction. Bagged Trees achieved the highest performance, with over 99% accuracy across metrics, and a user-friendly GUI-enabled real-time risk assessment. The GUI interface designed for these models provided users with accessible, dynamic feedback, enhancing usability for real-time assessment. Analysis revealed notable correlations, including age with pregnancies (0.54) and BMI with skin thickness (0.39), suggesting key factors in diabetes risk. Findings show that Glucose, BMI, Blood Pressure, and Diabetes Pedigree Function emerge as the top influential features, across all models. Using fractional-order modeling, we simulated glucose, insulin, BMI, and blood pressure changes over time, showing that higher fractional orders aligned with increased response dynamics. Together, these methods offer a comprehensive view of diabetes risk factors in women. By combining machine learning’s predictive power with the detailed, time-sensitive insights of fractional-order modeling, this study highlights key risk indicators and deepens our understanding of diabetes dynamics, ultimately supporting more effective risk assessment and management strategies. This dual approach provides an enriched perspective on diabetes risk factors and offers improved strategies for prediction and management.