Diabetic Retinopathy Detection Through Iris Using Genetic Algorithm, Wavelet and Particle Swarm Optimization

This research proposes a method for diagnosing Diabetic Retinopathy by extracting texture or image features and classifying them using genetic algorithms and wavalet for texture feature extraction as well as algorithm optimization using PSO. The implementation of the algorithm is carried out with a configuration using 1 hidden layer which has 10 chromosomes and 1 crossover layer which has 5 chromosomes which is obtained from the number of diagnoses produced. Testing of 2 feature comparisons was carried out using 2 test scenarios, namely changes in maximum epoch and changes in type and level of decomposition. The testing scheme used is K-Fold Cross Validation with a K value of 10. 50 datasets are divided into 10 folds to produce 5 test data and 45 training data as well as using learning rate and momentum configuration scenarios with values between 0.1; 0.2; 0.3; 0.4; 0.5; 0.6; 0.7; 0.8; 0.9 with a limiting the epoch value to a maximum number of 1000 training time (epoch) or using a stopping condition. If the error value obtained is lower, the specified error value is 0.0001 and the results have been obtained that the use of the genetic algorithm and wavalet in the system is able to produce an accuracy value of 79.00% with an MAE value of 0.000112 which is obtained at the 429th epoch while at The use of genetic algorithms, wavalet and PSO obtained an accuracy value of 82.70% with an MAE value of 0.000103 obtained at the 477th epoch.