Diagnosing Diabetes Disease Based on Machine Learning Techniques

Machine learning has been popular in medicine for offering methods and assessing disease data. Gradually, machine learning algorithms are becoming critical in attaining early disease diagnosis. Standard datasets of different types of diseases have been utilized in medical research and applications, including liver disease, diabetes, chronic kidney disease, heart disease, breast cancer, brain tumors, and many more. This research reviewed numerous machine learning algorithms for diabetes illness prediction. Subsequently, four machine learning classification algorithms of k-Nearest Neighbor (k-NN), Decision Tree (DT), Random Forest (RF), and XgBoost (XB) are selected to find the best model that performs the diabetes diagnosis. The Pima Indians Diabetes Database (PIDD) dataset is used for testing the models, which includes 768 instances and 10 features, and the class has labels of 0 that indicate negative diabetes or 1 that indicates positive diabetes. Precision, accuracy, and recall criteria are used to evaluate the performance of the algorithms. The experimental results show that the XB is the most suitable algorithm for building a diabetes diagnostic model. The XB model has the highest accuracy of 83.72% in the data split of 70% training and 30% testing. It has achieved an average accuracy of 81.20% as compared with the k-NN (74.02%), DT (77.71%), and RF (77.16%).