Abstract:

The prediction and early diagnosis of diabetes play a crucial role in managing the disease and preventing complications. This study presents a comparative analysis of the Support Vector Machine (SVM) classifier and the K-Nearest Neighbor (KNN) algorithm for diabetic prediction. Early detection using machine learning models can significantly improve healthcare outcomes. Existing methods often suffer from limitations such as poor prediction accuracy, long training times, and challenges in handling imbalanced datasets. These drawbacks can impact the effectiveness of diabetic prediction models in real-world applications, where quick and accurate decision-making is essential. The proposed framework, Prediction and Early Diagnosis of Diabetes Using SVM (PEDD-SVM), addresses these issues by leveraging the strengths of the SVM classifier. The method incorporates preprocessing steps like normalization and feature selection to improve data quality. It also applies kernel tricks for better decision boundaries, ensuring higher accuracy and generalization. This innovative approach aims to outperform traditional models by enhancing predictive accuracy and reducing computational complexity. By utilizing PEDD-SVM, healthcare professionals can identify diabetic patients at an early stage, allowing for timely intervention and better management of the disease. This method offers a more reliable and efficient tool compared to conventional algorithms, especially when working with large and complex datasets. The results indicate that PEDD-SVM provides superior classification performance, with higher accuracy and faster prediction times when compared to KNN. The findings demonstrate that the proposed model is effective in addressing the issues of imbalanced datasets and prediction latency, making it a promising solution for diabetes prediction.