Comparative Analysis of Predicting Diabetes in Senior Age Group Based on Machine Learning Models

The study investigated the contributing factors to risk for diabetes among older populations and their impact on the onset of diabetes. Analyzing a dataset from Kaggle comprising 18, 100 older adults, the study revealed significant associations between diabetes and age, Body Mass Index (BMI), HbA1c levels, blood sugar levels, high blood pressure, heart disease, and smoking history. Comparison of logistic regression, random forest, and K-nearest neighbor (KNN) prediction models demonstrated that the random forest model exhibited superior performance, with a ROC-AUC value of 0.88, an out-of-pocket error rate of 7.58%, a sensitivity of 99%, and a specificity of 92.3%. This outperformed both the logistic regression (ROC-AUC value 0.85) and KNN (ROC-AUC value 0.80) models. The study indicated that the random forest model is advantageous for processing nonlinear data and multi-variable interactions, making it suitable for predicting diabetes risk in the elderly. Future research could incorporate lifestyle factors to enhance prediction accuracy.