Assessing SVM and Logistic Regression Models for Live Birth Prediction in IVF: A Barbadian Case Study

Steven Cumberbatch, Adrian Als, Peter Chami, Juliet Skinner

The success rates of in vitro fertilization (IVF) have significantly improved over recent decades due to advancements in both clinical practice and biomedical technologies. Clinicians rely on the analysis of large volumes of patient data to inform treatment decisions. Aggregated longitudinal data from multiple patients may reveal latent patterns that can further enhance IVF outcomes. In this study, three machine learning models — Linear Kernel SVM, RBF-Kernel SVM and Logistic Regression — were developed and implemented to predict live-births from IVF clinical and demographic data, and their performances were compared. Results show that the linear SVM achieved the highest global discrimination (ROC-AUC = 0.72) and the strongest cross-validated F1-score (0.56). Logistic regression followed closely in global discrimination (ROC-AUC = 0.69), but its cross-validation recall for the minority class was notably low (0.26). The RBF SVM demonstrated a higher recall for the minority class compared to the linear SVM (0.45 vs 0.36), yet its overall discriminative performance was weaker, as reflected by a lower ROC-AUC of 0.63. This research serves as an initial exploration of machine learning applications in IVF within developing countries in the Eastern Caribbean, such as Barbados. The findings may contribute to improved clinical decision-making, reduced treatment cycles, and lower healthcare costs in resource-constrained settings.

Full Text