How to improve a 5-class Diabetic Retinopathy model (APTOS 2019) – Mixed predictions across classes[P]

Hi everyone, I'm a final-year Computer Engineering student building a Flask-based AI Diabetic Retinopathy Detection system. The web application itself is complete with patient management, authentication, dashboard, PDF report generation, prediction history, and AI inference. The only issue I'm facing is with the AI model. I'm using a 5-class Diabetic Retinopathy classifier trained on the APTOS 2019 dataset. Classes: No DR Mild Moderate Severe Proliferative DR The model predicts all five classes, but the pre