Diabetic Retinopathy (DR) is an eye disease that is more common in people with diabetes. When fluids, such as blood, leak into the retina, they weaken the retina's small blood vessels and eventually lead to blindness. Because of this, the retinal tissue expands, which impairs vision. Blindness from DR is permanent if it is not treated in the initial stages. Early detection through screening programs and referral to therapy is crucial for maintaining eyesight in DR patients. Retinal fundus images show microaneurysms, hemorrhages, hard exudates, and soft exudates. Each of these anomalies is extracted separately using standard DR detection methods. Predictions based on a single such anomaly become less reliable when there are numerous in existence. So, to categorize all the anomalies, a unique combination class was required. The ultimate purpose of our research is to create a machine-learning technique for DR prediction that can classify these anomalies into one of sixteen groups. Fundus image analysis using Merged Enhanced Green and Value Color Planes (MEGVCP), feature extraction using Speeded Up Robust Features (SURF), and DR classification using a Probabilistic Neural Network (PNN) are all components of the recommended approach. The PNN classifier has a success rate of 98%. The proposed method aids in speeding up diagnostics by assisting ophthalmologists in spotting anomalies either individually or in combination.