Early Prediction of Diabetes based on the Weight Enhanced XGBoost Model of AHP Clinical Guidelines

Abstract

Aiming at the problems of the traditional machine learning model in the early prediction of diabetes, such as insufficient interpretability, lack of clinical basis, and limited clinical applicability, this paper proposes a clinical interpretable intelligent prediction model that combines analytic hierarchy process (AHP) and extreme gradient lifting (XGBoost). According to the clinical diagnosis and treatment guidelines for diabetes, this method converts medical prior knowledge into quantifiable feature weights through AHP, and embeds the feature space to realize the deep coupling of clinical logic and data-driven learning, so as to improve the prediction accuracy and decision interpretability of the model. The comparative experiment on PIMA Indian diabetes dataset shows that the accuracy of the proposed model is 89.56%, and the AUC value is 0.943. All performance indicators are better than logistic regression, support vector machine, random forest and single XGBoost model. This model can not only accurately identify the risk of diabetes, but also its decision-making process conforms to the clinical cognitive law, and can provide reliable technical solutions for early screening of diabetes, early warning of high-risk groups and clinical auxiliary decision-making.