Design of a Multi-Model Diabetes Risk Prediction System for Clinical Application

Early diabetes prediction is significant for reducing the medical burden. This study aims to build a high-precision diabetes prediction model using machine learning methods to provide a reliable tool for clinical screening. Based on the Kaggle diabetes prediction dataset, which contains 100,000 patient records, four algorithms of random forest (RF), logistic regression (LR), support vector machine (SVM), and decision tree (DT) were used in this study. It adopted feature engineering such as gender coding, smoking history processing, and model optimization like default parameter comparison, grid search and 5-fold cross-validation. The prediction model is constructed. The results showed that the RF model had the best overall performance (accuracy 96.4%, F1 85.7%), SVM had the best recall rate (89.1%), and LR had high accuracy (95.3%) and interpretability. In comparison, DT achieved 94.6% accuracy with millisecond-level inference speed. Characteristic analysis further confirmed that blood glucose level, Body Mass Index (BMI), and age were the key predictors. This study can provide algorithm support for the hierarchical diagnosis and treatment of diabetes and optimize the early screening process.