Empirical Analysis of Diabetes Prediction Using Machine Learning Techniques

Abstract

In the world, there are many deadliest diseases, and diabetes is one of them. About 422 million people worldwide are suffering from this dangerous disease, and every year 1.6 million deaths are because of it. With each passing year, the number of cases is also steadily increasing. Even in India, diabetes is a growing health concern and requires great attention. To prevent diabetes and all its related problems, it is necessary to predict the earlier stages so that the treatment is done timely. Being motivated by this, in this paper, empirical analysis has been done on prediction of diabetes using machine learning techniques namely K-Nearest Neighbors, Support Vector Machine, Extra Tree, Random Forest, Naive Bayes, Bagged Decision Tree, Adaptive Boosting, Stochastic Gradient Boosting, and MLP Classifier. The results were validated on the standard PIMA Indian diabetes dataset, which is publicly available. We have evaluated the models based on their accuracy, sensitivity, precision, specificity, and F1 Score measures. It was observed that the K- Nearest Neighbors algorithm gave the best accuracy in comparison to the other aforesaid machine learning techniques.