چكيده به لاتين
Diabetes is a chronic disease that, along with its complications, is recognized as an important threat to public health. This disease leads to an increase in glucose concentration in the blood, which ultimately increases the possibility of complications such as eye problems, kidney, nervous, and heart failure in a diabetic person. Therefore, using a tool such as data mining that can effectively increase the accuracy of diagnosis and prediction of these complications seems vital. In the current research, using a data set related to patients with type 2 diabetes from Imam Khomeini Hospital, some complications such as retinopathy, nephropathy, neuropathy, and coronary artery disease caused by diabetes have been predicted. In this regard, to select the best factors causing the mentioned complications, three approaches of recursive feature elimination (RFE), Fuzzy DEMATEL as a decision-making technique, and Fuzzy Cognitive Maps (FCM) based on the neural network have been used for feature selection. The following algorithms include decision tree, Naive Bayes, K nearest neighbor, logistic regression, linear discriminant analysis, multi-layer perceptron neural network models, and combined models of bagging, boosting, voting, and stacking have been used for prediction. In the end, important predictor factors have been introduced for each of the complications, and the AUC of prediction has been obtained. The highest AUC obtained for each complication is retinopathy: 77.9%, nephropathy: 95.7%, neuropathy: 79.3%, and coronary artery disease: 80.9%.
