Prediction of diabetic retinopathy using health records with machine learning classifiers and data Science

Abstract

Diabetes is a rapidly spreading disease. It occurs when the pancreas produces insufficient insulin or the body cannot utilise it effectively. Diabetic retinopathy (DR) and blindness are two major issues for diabetics. Diabetes patients increase the amount of data collected about DR. To extract important information and undiscovered knowledge from data, data mining techniques are required. DM is necessary in DR to improve society's health. The study focuses on the early detection of diabetic retinopathy using patient information. DM approaches are used to extract information from these numeric records. The dataset was used to forecast DR using logistic regression, KNN, SVM, bagged tree, and boosted tree classifiers. Two cross-validations are used to find the best features and avoid overfitting. The dataset includes 900 diabetes patients. The boosted tree produced the best classification accuracy (90.1%) with 10% hold-out validation. KNN also achieved 88.9% accuracy, which is impressive. As a result, the research suggests that bagged trees and KNN are good classifiers for DR.