Neonatal health prediction is essential for timely intervention and improved outcomes. Accurate and early prediction of fetal health plays a crucial role in reducing neonatal complications. This study proposes a hybrid ensemble model framework combining TabNet and XGBoost models for classifying fetal conditions as either normal or abnormal using cardiotocographic (CTG) data. Clinically significant features such as maternal cortisol levels, sleep habits, placental position, fetal movement variability, and pH fluctuations are integrated to enhance model precision. The Synthetic Minority Over-Sampling Technique (SMOTE) addresses class imbalance, and soft voting merges outputs from both models to improve predictive strength.The proposed ensemble model achieves 93.90% accuracy, 90.43% recall, and an F1-score of 86.74% in identifying abnormal cases.Feature importance analysis reveals that placental location and maternal cortisol levels are among the most influential predictors in assessing fetal distress. The model demonstrates high interpretability and strong generalizability, offering a practical tool for clinical decision-making. Compared to prior methods, the proposed ensemble model surpasses prior techniques by delivering a dependable, interpretable, and efficient solution for early fetal health evaluation, making it ideal for real-time application in obstetric monitoring environments.