Diabetes Prediction System Based on Iridology Using Machine Learning

Abstract

Diabetes is a disease whose initial symptoms are often undetectable. As a result, many cases of diabetes are not detected early. Iridology can be an alternative to detect diabetes early. This method can reveal the state of the organ in the body before the appearance of symptoms of a disease. In this paper, a diabetes prediction system based on iridology or through iris images was constructed using machine learning. Machine learning used to simplify the detection process. The developed system consists of eye image acquisition instruments and image processing algorithms. Iris images were captured using Camera Iriscope Iris Analyzer Iridology. The GLCM (Gray Level Co-Occurrence Matrix) method is used for feature extraction processes to obtaining texture characteristics in the image. The kNN (k Nearest Neighbor) method are used to classify diabetic and non-diabetic classes. The classification results are then validated by using the k-fold cross-validation method and evaluated by using the confusion matrix. Two subject groups were evaluated: one was 16 subjects non-diabetic and 11 subjects diabetic. The results show that the accuracy is 85.6%, false-positive rate (FPR) is 11.07%, false-negative rate (FNR) 20.40%, specificity 0.889, and sensitivity 0.796.