Enhancing Eye Diseases Classification Using Imbalance Training & Machine Learning

Abstract

This research aims to evaluate the effectiveness of various machine learning algorithms in classifying eye diseases based on retinal images. The dataset comprises four categories of eye diseases: Cataract, Diabetic Retinopathy, Glaucoma, and Normal. The feature extraction method employed a transfer learning approach using ResNet50, followed by SMOTE for data balancing, PCA for dimensionality reduction, and normalization for scaling data consistently. Eleven machine learning models were evaluated, including basic algorithms, ensemble methods, and neural networks. The evaluation utilized metrics such as accuracy, precision, recall, and F1-score. K-Fold Cross Validation is also employed to observe all models' generalisation. The results revealed that the XGBoost algorithm achieved the highest performance with an accuracy of 92.03%, followed by LightGBM 91.88% and MLP 91.50%. K-Fold Validation also improved the MLP performance, which achieved an average accuracy of 91.94% with a standard deviation of 0.0178. This study successfully enhanced classification accuracy compared to previous studies and shows significant potential for clinical applications in resource-limited environments.This research aims to evaluate the effectiveness of various machine learning algorithms in classifying eye diseases based on retinal images. The dataset comprises four categories of eye diseases: Cataract, Diabetic Retinopathy, Glaucoma, and Normal. The feature extraction method employed a transfer learning approach using ResNet50, followed by SMOTE for data balancing, PCA for dimensionality reduction, and normalization for scaling data consistently. Eleven machine learning models were evaluated, including basic algorithms, ensemble methods, and neural networks. The evaluation utilized metrics such as accuracy, precision, recall, and F1-score. K-Fold Cross Validation is also employed to observe all models' generalisation. The results revealed that the XGBoost algorithm achieved the highest performance with an accuracy of 92.03%, followed by LightGBM 91.88% and MLP 91.50%. K-Fold Validation also improved the MLP performance, which achieved an average accuracy of 91.94% with a standard deviation of 0.0178. This study successfully enhanced classification accuracy compared to previous studies and shows significant potential for clinical applications in resource-limited environments.