Article
Predicting Diabetic Retinopathy and Nephropathy Complications Using Machine Learning Techniques
Diabetes and its complications, particularly Diabetic Retinopathy (DR) and Diabetic Nephropathy (DN), pose significant challenges to global healthcare systems, requiring accurate predictive models for early diagnosis and intervention. Traditional approaches often underperform due to imbalanced datasets and complex feature interactions. Publicly available datasets, including structured diabetic clinical data and the APTOS 2019 retinal fundus images, were utilized. Preprocessing involved KNN imputer for missing values, outlier detection and handling, MinMax scaling, and SMOTE oversampling to balance the data. Multiple machine learning models were implemented for classification, including variants of Logistic Regression, Random Forest, XGBoost, LightGBM, CatBoost, Multi-Layer Perceptron, as well as hybrid ensemble approaches. Advanced models were further developed, comprising StackingClassifier, Ensemble-ofEnsembles, and image classification architectures including ResNet50, DenseNet121, Xception, NasNetLarge, and an ensemble of Xception + DenseNet121. Evaluation metrics included Accuracy, Precision, Recall, F1-Score, ROC-AUC, RMSE, and LogLoss. DenseNet121 achieved the highest classification performance with 99.6% Accuracy, while StackingClassifier Oversampled reached 99.9% Accuracy for nephropathy, and LightGBM OverSampled attained 99.6% Accuracy for retinopathy. Explainable AI techniques, including LIME, SHAP, and Grad-CAM, provided model interpretability, and a Flask framework-based interface enabled user-friendly prediction deployment, ensuring accurate, transparent, and actionable clinical decision support.
Full Text Attachment
