Gestational diabetes (GDM) - a condition during pregnancy characterized by high blood sugar and lack of insulin response - raises not only the risk of developing Type 2 diabetes after pregnancy, but also the risk of complications for the fetus. While efforts are in place to identify pregnancies that are the most at risk for developing gestational diabetes, the current methods of risk assessment fail to identify both new variables that could correlate with risk of gestational diabetes and nonlinear relationships between these variables. In this study, Li et. al. attempted to create a machine learning model that could identify the variables involved in the overall risk of GDM development and the specific adverse maternal/fetal outcomes that they are correlated to. The machine learning model was trained on a GDM dataset that compromised of 1854 records of GDM patients: clinical and demographic information was noted, "adverse outcomes" like stillbirths/low newborn birth weight/prematurity were all defined, and patients with missing data were either excluded from the dataset or had any missing values deduced via statistical analysis of other patients' values for that category. Variables that were highly correlated were also removed to account for "redundancy." After finally processing all the data, a variety of algorithms were then used to create predictive machine learning algorithms; the initial predictions from each of these algorithms were then taken and used to train a secondary model (chosen to be XGBoost) in order to enhance predictive accuracy. More algorithms - called ADASYN and SMOTE - were further used to increase accuracy, ADASYN by increasing samples from the adverse pregnancy outcome group in the initial GDM dataset (only 11.9% of pregnancies in the initial GDM dataset had resulted in adverse outcomes, meaning that the data was imbalanced), and SMOTE by accounting for the effects of oversampling. In the end, 1670 patients' data was used to train a machine-learning algorithm, of which 200 of them had adverse pregnancy outcomes. Moreover, the model was then "stacked" with secondary models and further algorithms to increase accuracy chances - this strategy was proven effective, as the accuracy of the stacked model was 0.856, a promising measure, and the stacked model excelled in positive predictive value, positive likelihood ratio, and specificity when compared to the individual, non-stacked models that were developed. When testing the effects of ADASYN vs SMOTE on the stacked model, it was found that the stacked model with ADASYN had a better performance on most metrics, including sensitivity and positive predictive value, while the stacked model without any ADASYN or SMOTE analysis had a higher sensitivty value but with lower precision and specificity; the SMOTE-trained stacked model had a lower performance across most metrics.