This document proposes a new method for detecting cataract using statistical features extracted from fundus images and classifying the severity using machine learning algorithms. The method involves pre-processing fundus images using adaptive histogram equalization, extracting statistical features like mean, entropy and area using thresholding, and classifying the features using K-means clustering and ANFIS to determine if the image shows normal, mild, or severe cataract. The method achieved high accuracy, sensitivity and specificity for cataract detection and classification.