Diabetic retinopathy affects millions of individuals globally. If untreated, this condition, mostly affects the retina of the eye, results in permanent blindness. Therefore, it's crucial to identify Diabetic retinopathy in early stage to protect patients from going blind. This article proposes a novel ML approach that uses the mutual information technique for selecting the optimized features. . The efforts are made at the feature extraction and feature selection stage to select optimized feature set for the classification. The dataset for diabetic retinopathy is initially subjected to the ML algorithms Nearest neighbor classifier (NNC ),Naive Bayes classifier(NBC),Decision Tree classifier(DTC).The Random forest classifier outperforms them all in accuracy with an average performance of 75% after best feature selection and 66% before mutual information technique