Diabetes is a chronic disorder, characterized by low insulin production and high blood sugar levels in people of all ages. Diabetes, if left untreated, may lead to a variety of ailments throughout the body parts. Diabetic Retinopathy (DR) is a symptomless eye disease caused due to diabetes, where vessels present in retina of the eye are destroyed and wall of the vessel becomes weak. It is very important to catch the signs of Diabetic Retinopathy before it becomes too serious. Prolonged Diabetic Retinopathy will lead to blindness if left untreated and after that it cannot be reversed.So it is verymuch crucial to detect the diabetic retinopathy in the initial stage. Many of the present automatic diagnostic approaches make use of the decesion from the clinical practitioner. So, an efficient Deep learning and Machine Learning based method to classify the grades of diabetic retinopathy by segmenting different retinal lesions is proposed in this work because Deep Learning (DL) does automated feature extraction and it produces more accurate and potentially useful findings, especially in medical imaging. The methodology used in this paper provides both multilesion segmentation and disease severity diagnosis using an ensemble framework which is fully automated and computationally efficient and hence this method can be potentially included in CAD(computer-aided diagnosis) tools used for clinical practice.