A Hybrid ResNet50-Random Forest Model for Accurate Eye Disease Classification

Early and accurate detection of eye diseases is essential for preventing vision loss. This paper proposes a hybrid model that combines ResNet50 for feature extraction and Random Forest (RF) for classification to improve the automated detection of retinal diseases, including Diabetic Retinopathy, Cataract, and Glaucoma. The dataset used includes 4217 retinal fundus images across four classes: Normal, Diabetic Retinopathy, Cataract, and Glaucoma. The model addresses challenges such as imbalanced datasets and complex decision boundaries. ResNet50 is utilized to extract deep features from the images, while Random Forest efficiently classifies these features with high accuracy. The proposed model achieved strong results, with Precision, Recall, Macro-F1, and Weighted-F1 scores all exceeding 92%, and a low Misclassification Rate of 7.96%. Additionally, explainable AI techniques like Grad-CAM were employed to enhance the model's interpretability, which is vital for its use in clinical settings. The hybrid ResNet50-RF model outperformed standalone models, demonstrating its effectiveness in eye disease classification. This approach offers a promising, reliable solution for automated retinal disease diagnosis and has significant potential for broader application in healthcare.