Improving Accuracy of Diabetes Prediction Using Machine Learning Techniques and Clinical Data

Diabetes is a chronic condition affecting millions of people globally, requiring early and accurate prediction to mitigate associated health risks. This study employed a Kaggle dataset with 96,148 records, preprocessed through normalization, encoding, and the Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance. Two machine-learning models, Random Forest (RF) and Support Vector Machine (SVM), were trained and evaluated. Without addressing class imbalance, RF achieved 96.91% accuracy, with a lower computational time than SVM. After balancing the dataset, RF outperformed SVM with an accuracy of 95.42%, recall of 95.42%, and Area Under the Curve of 0.97, demonstrating superior performance in terms of both accuracy and efficiency. These results highlight RF as an optimal model for diabetes prediction, offering reliability and computational efficiency and making it a suitable choice for real-world healthcare applications.