A Hybrid Feature Selection and Logistic Regression Approach for High-Accuracy Early Diabetes Detection

Early detection of diabetes holds great importance for patient welfare because this chronic disease exists as one of the most common worldwide. This research develops an innovative machine learning diagnostic approach which provides efficient early diagnosis of diabetes. Our proposed methodology starts with data preprocessing before employing HSFE which blends ANOVA statistical filtration with RFE model-driven feature reduction to identify important predictors including Age, Glucose, Insulin, Pregnancies, and Diabetes Pedigree Function. Through its hybrid strategy the method achieves dimension reduction and discovers five important features comprising Age, Glucose, Insulin, Pregnancies, and Diabetes Pedigree Function. The selected approach for modeling consists of a logistic regression model because it allows for both accurate interpretation and strong performance in binary classification tasks. The model demonstrates a high accuracy level of 97.84% backed by precision, recall and F1-score values exceeding 96% which confirms dependable and balanced functioning. The results confirm that the integration of statistically grounded and algorithmically optimized feature selection significantly enhances predictive accuracy while maintaining clinical relevance. This work contributes a scalable and interpretable diagnostic tool suitable for real-world healthcare deployment, addressing the need for accurate, early-stage detection of diabetes with minimal computational overhead.