Diabetic Retinopathy (DR) is one of the most common complications of diabetes and a major cause of blindness in the world. Screening is very important in the management and prevention of the diseases from getting worse. This paper aims at evaluating the effectiveness of CNN models including VGG16, VGG19, ResNet50, ResNet101, MobileNet, InceptionResNetV2, and DenseNet for the identification of Diabetic Retinopathy from retinal images. For the purpose of improving model interpretability, Explainable Artificial Intelligence (XAI) methods are applied, underlining the need to explain the decision-making of deep learning models. Feature importance analysis is performed in order to determine which of the retinal features have the most impact on the model. This work does not only give an insight into the biomarkers of Diabetic Retinopathy, but also helps clinicians in decision making for diagnosis and management of the condition. The combination of deep learning with Explainable AI (XAI) methods meets one of the most important requirements for AI in healthcare, namely, the ability of clinicians to understand the models they employ. Out of all the models, the custom built Deep CNN model named “DR NET V2” which was built using the Fast AI framework had the highest accuracy of 84.29%. This result has shown that the proposed DR NET V2 model is capable of detecting Diabetic Retinopathy with high accuracy, thus its applicability in real world scenarios. This work adds to the current research on creating accurate and explainable models for early identification and monitoring of Diabetic Retinopathy.