Identification of Diabetic retinopathy (DR) classification presents a challenge in medical imaging analysis since the differences between the various stages of the disease are subtle. Hence we have introduced a new machine learning pipeline which combines Convolutional Neural Networks (CNN) with Particle Swarm Optimization (PSO) and eXtreme Gradient Boosting (XGBOOST) to increase the accuracy of multi-class classification of DR severity levels. In particular, we compared this hybrid model to the standard Random Forest and SVM methods as well as to standard CNN and LSTM_RNN models against common performance metrics such as accuracy, recall, precision, F-score, and AUC. CN-PSO-XGBOOST well improved the above models, achieving great accuracy, high sensitivity, and specificity even in the challenging ‘Proliferate_DR’ and ‘Severe’ levels, and performed excellently in detecting the advanced stage DR. The findings of this study highlight the promise of combining sophisticated optimization and boosting techniques when building robust, accurate classifiers for medical diagnostics and set a precedent for the application of machine learning in healthcare moving forward.