Optimizing IVF Diagnosis Prediction: Evaluating Feature Relevance Using Machine Learning

Despite being an increasingly prevalent assisted reproductive technique, in-vitro fertilization (IVF) has inconsistent success rates. Accurately predicting IVF outcomes can enhance clinical decision-making, reduce patient uncertainty, and optimize treatment plans. This study applies machine learning-based hierarchical feature selection to assess the relevance of various predictive factors in IVF diagnosis. We employ Random Forest, AdaBoost, Recursive Partitioning and Regression Trees (RPART), and Artificial Neural Networks (ANN) to rank feature importance and improve classification performance. The dataset includes clinical and lifestyle variables such as smoking habits, alcohol consumption, surgical history, embryo type, sperm quality, and the number of embryos. Standardization techniques, class-weight balancing, and stratified sampling are applied to ensure robust model evaluation. Model efficacy is compared using performance indicators such as accuracy, precision, recall, F1-score, and the ROC curve. Results indicate that ANN achieves the highest accuracy (96%) and AUC (0.98), outperforming other models in distinguishing successful and failed IVF cases. Feature importance analysis highlights sperm quality and the number of embryos as the most influential factors, followed by lifestyle and medical history attributes. These findings provide valuable insights for developing more effective and interpretable IVF prediction models, aiding reproductive health specialists in personalized patient counseling and treatment optimization.