Diabetic retinopathy is a major metabolic disorder caused by elevated blood sugar. This study considers fundus images and introduces a stochastic neighbor embedding (SNE) feature extraction approach to reduce dimensionality and noise. An optimized deep belief network (O-DBN) classifier predicts disease severity. The cloud-enabled system improves sensitivity, specificity, F1-score, prediction time, and accuracy compared to existing systems. © 2023 Elsevier B.V., All rights reserved.