Using machine learning techniques for EARLY diagnosis of TYPE 2 diabetes

Objective: Diabetes, a chronic metabolic disorder, is a lifelong disease that occurs when the pancreas no longer produces enough insulin or when the body cannot effectively use the insulin it produces. In this study, the use of artificial intelligence technology to predict type 2 diabetes is addressed. The aim is to lighten the additional workload on healthcare systems due to early diagnosis, improve the quality of life of patients, and prevent them from experiencing economic difficulties due to high treatment costs. Method: We examined data from 126 patients obtained from private hospitals in the Mediterranean, Marmara, and Aegean geographical regions. We grouped patients diagnosed with Type 2 Diabetes by specialist physicians and control patients without a diabetes diagnosis. Logistic Regression (LR), Random Forest (RO), and Decision Trees were used to create models for predicting the group. AUC value, accuracy, sensitivity, specificity, precision, and F1 score were used to compare the performances of machine learning algorithms. Results: Complete blood count (hemogram) test results and sociodemographic characteristics were used for individuals with Type 2 diabetes and the control group. Logistic Regression, Random Forest, and Decision Tree techniques yield ROC AUC values of 92.6% (88.1-97.1), 73.6% (65.9-81.4), and 84.5% (78.3-90.7), respectively. The Logistic Regression algorithm achieved higher success in classification with a ROC AUC of 92.6% and 88.1-97.1 confidence interval compared to other machine learning algorithms. Conclusion: Although no new method proposal was made in this study, it provides a foundation for future similar studies and presents data for comparing the machine learning algorithm studied. The most successful method for predicting Type 2 diabetes was found to be Logistic Regression. Keywords: Type 2 diabetes mellitus, complete blood count, machine learning, artificial intelligence