Diabetic Retinopathy Prediction by Ensemble Learning Based on Biochemical and Physical Data.

Area of Science:

• Medical Informatics
• Machine Learning
• Ophthalmology

Background:

• Diabetic retinopathy is a leading cause of blindness in developed nations.
• Early detection and treatment of diabetic retinopathy are crucial.
• Predicting diabetic retinopathy from complex, small-sample datasets presents a challenge.

Purpose of the Study:

• To develop an accurate prediction model for diabetic retinopathy.
• To address the challenge of high-dimensional, small-sample datasets in prediction.
• To improve early screening and reduce diagnostic costs for diabetic retinopathy.

Main Methods:

• Proposed the XGB-Stacking model, integrating XGBoost and stacking ensemble methods.
• Utilized XGBIBS (Improved Backward Search Based on XGBoost) for feature selection and redundancy reduction.
• Employed Sel-Stacking (Select-Stacking) for optimal learner combination and model fusion.

Main Results:

• XGBIBS significantly enhanced prediction accuracy and feature reduction rates.
• The Sel-Stacking model demonstrated improved accuracy compared to single classifiers.
• The XGB-Stacking model effectively predicted diabetic retinopathy using biochemical and physical data.

Conclusions:

• The XGB-Stacking model shows outstanding performance for diabetic retinopathy prediction.
• This approach offers significant improvements in screening efficiency.
• The model has the potential to reduce the overall cost of diabetic retinopathy diagnosis.