Improvement of support vector machine for predicting diabetes mellitus with machine learning approach

The prevalence of diabetes is currently increasing worldwide, including in Indonesia, due to the increasing levels of stress and lack of physical activity that led to obesity and related complications such as hypertension. However, only about 25% of diabetes patients are aware of their condition. Therefore, this study aims to find an algorithm that can help predict with better accuracy using the diabetes mellitus dataset obtained from Kaggle. To obtain information about the accuracy level of diabetes diagnosis, the data will be processed using two methods, namely support vector machine and naive bayes. To obtain the most accurate results, we optimize each variant and parameter of every algorithm used. The best method in this study was produced by the support vector machine method with a radial basis function (RBF) kernel, which achieved an accuracy level of 98.25%, superior to the naive bayes method which obtained the highest accuracy of only 77.25%. Additionally, this study also applied the proposed method using the diabetes mellitus dataset from LAB01 DAT263x taken from the Kaggle website. The results of the experiment indicate that the suggested model outperforms other methods in terms of performance, with a tendency for high accuracy generated in every experiment for all datasets.