Iranian Journal of

---

Background: Diabetes is such diseases that need high quality beside prevention such as correctly predict fluctuations in blood glucose levels. The main complications of the disease can be anesthesia, coma and even death. Today, in these patients, the correct dose of insulin determined based on experience or doctors knowledge, and interact between the patients and physician, although there is an inevitable human errors.

Methods: In this study based on applied method, 124 patients and 188 healthy subjects based on 12 features by Random Selection, Who had been referred to Research Center for Diabetic in Sabzevar university of Medical Science since 2006 to 2011 were studied. The proposed system has several subsystems, such as evolutionary algorithms (BPS 1) to select the most effective features, Data Mining Algorithms (SVM 2) to detect and classify the features from the non-effective features. Adaptive Neuro fuzzy systems (ANFIS 3) to estimate learn and adaptation in order to correctly predict have been used.

Results: In this study, we try to use artificial intelligence systems to determine the correct dose of insulin for diabetics. The proposed system combines the best attributes in the database in the form the interaction was able to achieve high accuracy with the lowest error. The proposed system based on best features in the database in the interaction form was able to achieve high accuracy with the lowest error. The proposed system in the form of composition and interaction with the subsystem was able to achieve carefully 84.1% in specificity, 91% in sensitivity and 92.9% in accuracy.

Conclusion: In this research, due to the importance of correct and timely determination of insulin for diabetics, a new method based on the combination of intelligent systems is presented. Thus, the results obtained in previous articles and studies provide significantly improved.

---

Background: On time diabetes diagnosis dramatically reduces the many injuries and damage in the community. Diabetes is a disease that requires a lot of care in addition to prevention, such as prediction the correct level of blood sugar fluctuations. The most important complications of such disease are anesthesia, coma and even death at final. Today, in these patients, determining the correct dose of insulin is based on the experience and knowledge of physicians along with the interaction of patients with them, although human error is inevitable.
Methods: This study includes 124 patients and 188 healthy suspects were examined based on 21 features which hold by 7 features for diagnosis and 14 features for predicting insulin dose. The proposed system was presented to identify or diagnose the disease at first, and finally the correct doses of insulin for patients have been determine. The proposed system has two stages (which include diagnosis and prediction) and several subsystems. In the diagnosis phase, some sub systems such as the Fuzzy system for the purpose of accurately estimating the disease progression in patients and the decision tree (DT) for the preparation of rules in the fuzzy system (the process of mapping the attribute space (individuals) to the output (the diagnostic result)) have used. Also, in the prediction phase of insulin dose, the BPSO algorithms are used to select the best features. Classification algorithms (SVMs) are used to categorize effective to non-effective and adaptive artificial neuropsychological (ANFIS) systems for ultimate patient prediction have used.
Results: The proposed system, based on the best features in the provided data base in the form of the combination and interaction, succeeded to achieve a 95.1% precision, of course due to comparing by other commonly used methods and its performance the proposed method have high precision.
Conclusion: The results were significantly improved compared to previous studies. Also, in comparison with the results of physicians, it is indicative of good performance in predicting the accuracy of the time series of glucose concentration because the proposed system succeeded in predicting blood sugar levels for up to 48 hours.