Early Diagnosis of Diabetic Retinopathy through Optimization of Convolutional Neural Network Hyperparameters using Genetic Algorithm

Diabetes is a chronic disease in the form of a metabolic disorder characterized by sugar levels that exceed normal limits. Indonesia is ranked 7 from 10 countries with the highest number of sufferers. Diabetes Mellitus is the main cause of blindness, kidney failure, heart attacks, strokes and lower limb amputations. One of the serious complications of diabetes mellitus is diabetic retinopathy (RD), which is the main cause of blindness. Examination costs are expensive and there are a few of professional experts to detect RD disease early. To overcome this problem, the use of intelligent systems such as Convolutional Neural Network (CNN) has been widely used. In this study, we optimized the CNN model for early diagnosis of RD using a Genetic Algorithm (GA) so as to optimize the performance of the Diabetic Retinopathy classification model. This method allows finding the optimal combination of the number of convolution layers and filters in each CNN layer and increases accuracy. GA is part of a metaheuristic algorithm that is used to find optimal or near-optimal solutions to complex problems. GA facilitates model evolution by creating variations in CNN architecture. This process reduces the risk of overfitting and improves model performance on test data. This research is using the APTOS (Asia Pacific Tele-Ophthalmology Society) 2019 Blindness Detection dataset, which contains 3662 training images. The performance of the proposed method is evaluated using classification accuracy, precision, recall, f1-score matrices and combined in a confusion matrix. The results show that the CNN model achieves an accuracy is 93.01% and the CNN model that has been optimized with GA shows increased results reaching an accuracy is 9 7. 4 5%.