This document presents a study on developing an automated severity classification model for diabetic retinopathy using deep learning techniques. The proposed model uses a modified DenseNet169 architecture with a Convolutional Block Attention Module to classify retinal images into different severity categories of diabetic retinopathy. The model was trained on the Kaggle Asia Pacific Tele-Ophthalmology Society dataset and achieved state-of-the-art performance, accurately classifying 82% of images for severity grading. The lightweight model requires less time and complexity compared to other methods, making it suitable for automated diagnosis of diabetic retinopathy severity.