Diabetic retinopathy is a complication of diabetes that impacts vision. It arises from damage to blood vessels within the eye’s light-sensitive tissue. Initially, diabetic retinopathy may manifest without symptoms or with only minor vision issues. However, it can ultimately result in blindness. Machine learning methods within a Computer-Aided Diagnosis (CAD) system can assist specialists in diagnosing and evaluating the progression of DR. Thus, this work aims to assess and compare the effectiveness of different machine learning models in classifying the severity level of retinopathy in diabetic patients, utilizing the quantity and area of lesions in the retina. We test each binary and multi-class classification algorithm and calculate the F1-score, recall, and precision metrics. In the binary classification, our best result achieved a recall of 91.2%. The multi-class results yielded satisfactory outcomes. As for the multi-class classification, the “mild” class was accurately detected, with the MLP being the most balanced model and achieving a recall rate of 84.1%. However, the “moderate” class exhibited slightly lower metrics than the “mild” class. On the other hand, the “severe” class demonstrated good detection despite being the lowest among all others.