Razi Journal of Medical Sciences

Background: Diabetes ever-increasing prevalence and the heavy burdens of controlling and treatment of the disease on people and the country have turned to be greatest challenges for governmental and healthcare authorities. Therefore, the disease prevention takes top priority and to do so the only possible way is detecting the effective parameters and controlling them. This study is about to foresee diabetes rates on the basis of some effective factors and using the artificial neural network. 

Methods: This study is conducted in 2014 by using R and SPSS software on 13423 participants of the study evaluation of risk factors of non-communicable diseases which was run in 2007. All the participants were older than 25 and with uncontrolled diabetes. A three-layer artificial neural network was used to evaluate the data, and to choose the best model the area under the ROC curve (AURC) and the prediction accuracy were applied. In this model both applied activation functions were Sigmoid.

Results: The three-layer artificial neural network with the architecture of (53:20:2) was identified as the best  model as the area under the ROC curve (AURC), the training prediction accuracy, and the test prediction accuracy were 72.7%, 92%, and 91.6% efficient, respectively.

Conclusion: Since in artificial neural network there is no need for common assumption of classic statistical methods and its high prediction accuracy (53:20:2) it is highly recommended to apply this model in predicting diabetes.and factors affecting it, that requires a separate study and research.