Diabetic Retinopathy (DR) is a serious retinal condition resulting from diabetes. In contemporary society, it significantly contributes to blindness and visual impairments. Therefore, the early prediction of DR is essential via appropriate screening tests administered periodically for people with diabetes. Moreover, the automated identification and categorization of diabetic retinopathy by sophisticated computational methods would facilitate early diagnosis and prompt treatment. The ensemble-based machine learning model demonstrates superior categorization of DR regarding sensitivity and accuracy compared to individual machine learning techniques. The research presents an automated DR classification system using preprocessing, feature selection, and machine learning classifiers for classification. The necessary characteristics may be chosen via feature selection techniques such as Chi-Square, ANOVA, and Improved Lasso. The tests are conducted using many benchmark datasets, including MESSIDOR, IDRID, and APTOS. The research indicates that the proposed model has effectively chosen features and minimized duplication in DR datasets, achieving a high-performance ratio. Furthermore, the classification accuracy has improved by 3–5% for the MESSIDOR, IDRID, and APTOS datasets, respectively.