Comparison the power of machine learning methods against traditional statistical approaches in predicting gestational diabetes: a study protocol

Background Gestational diabetes (GDM) is linked to numerous negative pregnancy results for both the mother and the infant. Accurate predictions of GDM trends are essential for public health. Machine learning models have become effective instruments for disease prediction, showing enhanced performance compared to conventional techniques by identifying complex temporal relationships and nonlinear trends within health data. Consequently, we aimed to evaluate the power of machine learning compared to conventional statistical techniques in predicting GDM. Methods The birth records of all expectant mothers who were admitted from January 2020 to January 2022 to one of the primary referral tertiary centers situated in Bandar Abbas, Iran, will be incorporated. We will employ 2 methods for analysis. In the initial stage, we will employ conventional analytical techniques. The chi-square test will evaluate the relationship between categorical variables and GDM. Bivariate logistic regression will be performed to analyze the risk factors for GDM, estimating crude odds ratios (cORs) along with their 95% confidence intervals (CIs). At the second level, we will employ machine learning techniques to forecast GDM. The input data will be utilized in eight machine learning models. To assess the diagnostic capability of each model, the area under the receiver operating characteristic curve (ROC AUC), accuracy, precision, sensitivity, specificity, and F1 score will be calculated. Discussion The findings of this research show the hospital's existing incidence rates for gestational diabetes. Recognizing the risk factors for gestational diabetes forms the foundation for strategizing preventive actions against it.