A Hybrid Machine Learning Strategy Assisted Diabetic Retinopathy Detection based on Retinal Images

Retinopathy is a serious disease occurred over the retinal area of the eye, in which it is mainly raised based on the Diabetic disease. This kind of retinal disease is named as diabetic retinopathy; it may cause the permanent disorder of an eye. This retinopathy disease affects the blood flow ratio of the retinal veins and cause the blindness to the people as well as it is caused by the irregular blood flow over the veins. This kind of diabetic retinopathy disease results from the damage to the retinal back portion, in which it is caused due to the propensity to the retina. An improper maintenance of Blood Sugar level leads to such risk cases and the diabetic retinopathy can easily be identified by some earlier symptoms such as appearance of floaters, decreased visual acuity, redness, yellow, and orange colors and poor color perception. These are all the common symptoms raised on earlier stages of diabetic retinopathy disease, in which it is recoverable but in case of poor consideration regarding such causes leads to permanent blindness. At the low end of the spectrum, the condition can be managed with careful control of one's diabetes. For more difficult cases, surgery or laser resurfacing may be required. In this paper, a digital image processing logic is utilized to process the retinal images and classify the normal and severe states in clear manner with respect to machine learning principles. This paper introduced a new machine learning strategy by means of combining two powerful machine learning algorithms such as Random Forest Classifier and the AdaBoost Classifier, in which it is integrated together to make a hybrid algorithm called Hybrid Retinal Disease Detection Logic (HRDDL). This proposed approach of HRDDL assures the logic of identifying the retinopathy diseases in clear manner with proper classification logics. The digital retinal image dataset downloaded from Kaggle database is utilized to prove the efficiency of the proposed approach and the resulting scenario is cross-validated with traditional Random Forest Classifier to prove the proposed HRDDL classification accuracy. This paper assures the HRDDL accuracy over prediction of diabetic retinopathy on earlier stages as well as the resulting section shows the clear proof for the identification of disease and the accuracy ratio. The proposed approach of HRDDL provides the accuracy range of 92.5% in results as well as this will be cross-validated with the classical Random Forest classifier to prove the efficiency well.