Enhanced SVM Optimization for Diabetes Prediction Using Particle Swarm and Differential Evolution Techniques

Abstract

In this study, the Pima Indians Diabetes dataset was used to explore an enhanced hybrid optimization framework for a Support Vector Machine (SVM) for diabetes prediction. Owing to the complexity of the dataset, a comprehensive approach for feature selection and hyperparameter tuning is essential. Particle Swarm Optimization (PSO) was employed for feature selection, whereas hyperparameters were fine-tuned using a combination of grid search and Differential Evolution (DE). In addition, k-fold cross-validation was applied to ensure robust performance across different data splits. Ensemble methods, including boosting techniques, have been investigated to further improve classification accuracy. The optimized model achieved an accuracy of 78.88%, which is a significant improvement over the 74.89% accuracy of the non-optimized nonlinear SVM. While the optimized model showed better results, the chapter also discusses potential future advancements in medical data analytics, such as advanced ensemble methods and improved data preprocessing techniques.