Filter Gabor and SMOTE Method-Based Convolutional Neural Network for Diabetic Retinopathy Classification

Diabetic Retinopathy is a disorder of the retina that occurs in people with diabetes and has an increasing risk with age. To prevent the worst risk of blindness, treatment and early detection of eye fundus images are needed. The problem in the image classification is how the model created can have a better performance value. This research aims to classify fundus images with Gabor and SMOTE filters to improve model performance by improving diabetic retinopathy retinal fundus images. The algorithm employed is deep learning, which is a problem-solving method for computer learning systems that use the concept of hierarchy. Among deep learning algorithms for solving detection or identification problems is the convolutional neural network (CNN). In addition, the Gabor Filter is also used to extract retinal fundus image features and SMOTE to overcome unbalanced datasets. The study results showed that Gabor and SMOTE filters could significantly increase the accuracy value from 40% to 78%. The increase in performance value was influenced by preprocessing, namely image resizing, converting to grayscale, and setting Gabor filter parameters.