Early Prediction of Gestational Diabetes Using ‎Decision Tree and Artificial Neural Network Algorithms

Introduction: Gestational diabetes is associated with many short-term and long-term complications in mothers and newborns; hence, the detection of its risk factors can contribute to the timely diagnosis and prevention of relevant complications. The present study aimed to design and compare Gestational diabetes mellitus (GDM) prediction models using artificial intelligence algorithms. Materials and Methods: In this study, Decision Tree and Artificial Neural Network algorithms were used to predict GDM. The research population encompassed 1270 pregnant women referred for primary care at urban healthcare centers in Ahvaz, of whom 816 persons were healthy, and 454 individuals were diagnosed with GDM. To evaluate the effectiveness of the GDM prediction models, their sensitivity, specificity, precision, and accuracy were calculated and compared. Finally, the AdaBoost classification algorithm was used to boost the two proposed models. Results: Following the Principal Component Analysis (PCA), nine cases were selected for primary modeling. In the Artificial Neural Network model, the area under the ROC curve and sensitivity were 83.2 and 85.1%, respectively, and the area under the ROC curve and sensitivity for the Decision Tree model were 0.826 and 84%, respectively. After removing variables with lower weights and reinforcing the proposed model, the level under the rock curve and sensitivity increased by 0.861 and 92.1%, respectively. In this regard, fasting blood sugar at the first pregnancy visit, history of gestational diabetes in previous pregnancies, body mass index, mothers’ age, and family history of diabetes had the highest accuracy in predicting GDM. Conclusion: The findings of this study indicate that artificial intelligence algorithms are accurate and effective for the early prediction of gestational diabetes.