Comparative Study of Early Diabetes Risk Stratification Based on Machine Learning Algorithms

Objective: This study aims to systematically compare the performance of multiple machine learning models in diabetes risk prediction and identify key risk factors, thereby providing data-driven decision support for early diabetes screening. Methods:Using the UCI Pima Indians Diabetes dataset, five models-logistic regression, K-nearest neighbors, support vector machine, decision tree, and random forest-were trained and evaluated. Model performance was comprehensively assessed via metrics including AUC-ROC, precision, and recall, with feature importance analysis employed to elucidate core diabetes risk factors.Results: The random forest model demonstrated superior performance across multiple metrics (AUC = 0.8167). Plasma glucose was consistently identified as the strongest predictor, with body mass index (BMI) and age also emerging as significant contributors. Conclusion: The random forest model exhibits robust performance and effective capture of feature interactions, making it well-suited for early diabetes prediction with considerable potential for clinical application.