Improved Models for Diabetes Prediction by Integrating PCA Technique

Abstract

Diabetes is an incurable, chronic disease indicated by hyperglycemia. It may lead to severe complications and is consequently one of the top ten causes of mortality in recent years. Fortunately, early detection can significantly aid in disease management. Based on their regular medical checkup results, people can make a preliminary evaluation of their risk of having the condition with the help of machine learning techniques. In this study, diabetes is predicted using Extreme Gradient Boosting (XGB), Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), Decision Tree (DT), and K-Nearest Neighbor Algorithm (KNN), ranked in descending order of accuracy. XGB comes out on top with an accuracy of 87.29%. The dimensionality reduction approach Principal Component Analysis (PCA) is conducted, resulting in a maximum accuracy gain of 1.27%. The study's enhanced diabetes prediction models allow people to more accurately gauge their risk of developing the condition. More potential patients would be notified to get checked, enhancing the disease's early detection rate.