Comprehensive Meta-Learning Approach for Predicting Gestational Diabetes Mellitus: A Comparative Analysis of Imputation Methods and Classifiers

Gestational Diabetes Mellitus (GDM) constitutes a health risk for everyone during pregnancy; as such, it requires proper early diagnosis and booster. This work offers a detailed analysis of machine learning approaches for GDM prediction, focusing on the interaction between imputation strategies and classification models. The study utilizes a robust benchmark dataset from Kaggle to evaluate the quality and reliability of classifiers, including Decision Stump, Decision Table, Bayes Net, and KNN classifiers, as well as SVM and imputation with both KNN and SVM. We compare the results using performance measures such as accuracy, precision, recall, F1-score, mean cross-entropy, ROC, and PRC scores. The results establish that KNN imputation performs appreciably better than SVM imputation, mainly in terms of prediction accuracy and more even-handed performance for almost all classifiers. Several models integrating KNN imputation with the sophisticated classifier accurately present the performance landscape, with classification accuracy at not less than 97%. Although some SVM-based models’ excellent performance shows higher predictive accuracy, they prove to possess greater complexity. This study's results show the importance of choosing the right classifier and imputing missing data when making machine learning models for GDM prediction. By handling the missing data and improving the classification methods, the study provides a direction for a large-scale, accurate, and efficient diagnostic approach, which may significantly improve women's health through prompt and accurate GDM identification.