Designing a Feature Selection Model to Enhance the Prediction of High-Risk Diabetic Patients using Machine Learning Techniques

This study aims to design a feature selection model to improve the accuracy of predicting high-risk diabetic patients using machine learning. Feature selection reduces the dataset’s complexity by removing irrelevant variables, thus enhancing model performance and reducingnoise. We applied statistical methods like the Chi-Squared test and feature importance analysis to identify the most significant features, such as HbA1c, BMI, age, and cholesterol levels. Seven machine learning modelsSVC, Decision Tree, KNN, Logistic Regression, Random Forest, AdaBoost, and Gradient Boostingwere trained with selected features. Performance evaluation indicated that Gradient Boosting achieved the highest accuracy of 97%. Cross-validation further refined model performance, with improvements noted in K Neighbors and Decision Tree classifiers when feature selection was applied. The study underscores the importance of feature selection in creating efficient, accurate models for diabetes prediction. By focusing on key features, the model accelerates training and offers better predictive capabilities, ultimately aiding in the early identification and management of high-risk diabetic patients, thereby enhancing clinical decision-making.