The main goal of this thesis is to analyze risk factors on fasting blood glucose and type 2 Diabetes with using data of Tehran Lipid and Glucose Study (TLGS). Initial results of TLGS show the high rate of metabolic disorders such as Diabetes. Diabetes is one of the costly diseases, in addition, one of the risk factors on cardiovascular diseases with effects on blood vessels. In TLGS, due to the practical limits, simultaneous effects of variables such as nutrition, physical activity, demography, anthropometric, clinical examination, medical records and drug consumption on blood glucose have not been analyzed. In this thesis, effect of all these variables on blood glucose are analyzed. Therefore, it is necessary to model relations between considered variables and blood glucose. One of the best techniques to model complex systems in various fields such as healthcare is data mining. In this thesis, for analyzing relations between variables, data mining models, both supervised and unsupervised, are applied. Before modeling, the data is cleaned and described using statistical tools. Because of capabilities of Artificial Neural Networks (A) in modeling nonlinear relations between variables, they are widely used in this research. Therefore, a neural network is proposed to predict fasting blood glucose. In addition, "Self Organizing Maps", as an unsupervised model for analyzing relations between variables, is employed. The work also consists the sensitivity analysis of the predicting model. To identify the main risk factors of Diabetes, a multilayer perceptron and a Logit model are proposed to diagnose Diabetes. Results of sensitivity analysis show that generally anthropometric variables (waist, hip, wrist and Body Mass Index) are found more important than others. ltr"
