Received January 20, 2026
Accepted February 09, 2026

Selecting the most viable embryo for transfer is a crucial step in IVF, significantly influencing treatment success. Despite technological advances, current embryo evaluation methods mainly based on morphology and time-lapse imaging have limited predictive power. IVF outcomes depend on multiple factors, including patient characteristics, endometrial receptivity, and semen quality. The study analyzed data from 375 IVF cycles conducted between 2013 and 2022, integrating morphokinetic, clinical, and semen-related parameters to predict live birth outcomes. Four machine learning models logistic regression, SVM, neural networks, and XGBoost were compared, with XGBoost showing the best and most stable performance (AUC > 0.6). After extensive hyperparameter optimization using Bayesian methods and synthetic data augmentation, the final XGBoost model achieved a high predictive accuracy (AUC = 0.927). Key predictive features included embryo class, timing of the second polar body extrusion, AMH concentration, female age, and BMI, leading to the creation of an application capable of estimating individualized live birth probabilities in IVF treatments. An AI-based assessment integrating clinical, embryological, and semen-related variables represents a promising approach to advancing personalized medicine. Further refinements and prospective validations are required to enhance prediction accuracy and prevent overestimation, ultimately optimizing cumulative live birth rates while reducing surplus embryo production.