Comparison of Gestational Diabetes Prediction Between Logistic Regression, Discriminant Analysis, Decision Tree and Artificial Neural Network Models

Background and Objectives: Gestational Diabetes Mellitus (GDM) is the most common metabolic disorder in pregnancy. In case of early detection, some of its complications can be prevented. The aim of this study was to investigate early prediction of GDM by logistic regression (LR), discriminant analysis (DA), decision tree (DT) and perceptron artificial neural network (ANN) and to compare these models.   Methods: The medical files of 420 pregnant women (2010-12) in Kermanshah health centers were evaluated using convenience sampling. Demographic data, pregnancy-related variables, lab tests results, and a diagnosis of GDM according to a fasting blood sugar level of 92 or more were collected from their files. After fitting the four models, the performance of the models was compared and according to the criteria of accuracy, sensitivity and specificity (based on the ROC curve), the superior model was introduced.   Results: Following the fitting of LR, DA, DT and perceptron ANN models, the following results were obtained. The accuracy of the above models was 0.81, 0.83, 0.78 and 0.83, respectively, the sensitivity of the models was 0.50, 0.63, 0.58 and 0.58, the specificity of the models was 0.96, 0.93, 0.87 and 0.94, and the area under the ROC curve was 0.86, 0.78, 0.73 and 0.87, respectively.   Conclusion: In predicting and categorizing the presence of GDM, the ANN model had a lower error rate and a higher area under the ROC curve compared to other models. It can be concluded that this model offers better predictions and is closer to reality than other models.