The cardiotocography (CTG) data are critical in ensuring that the health status of the fetus is assessed reliably in order to identify any adverse perinatal condition at an early stage. Manual interpretation of CTG recordings is, however, subjective and is prone to inter-observer variation. Here, this paper introduces a hybrid ensemble machine learning framework for fetal health classification using two publicly available CTG datasets. Using stratified 5-fold cross-validation, the baseline classifiers, ensemble strategies, and hybrid configurations that employ SMOTE-based imbalance correction, dimensionality reduction, and feature selection were systematically evaluated. Macro-F1 was employed as the primary performance metric. The experimental findings show that ensemble models consistently outperform standalone classifiers, with hybrid configurations being even more successful in recognizing minority classes. The Extra Trees classifier was optimized and scored 0.912 in the Fetal dataset and 0.989 in the CTG dataset with Macro-F1. Multi-class ROC analysis had a high ability to discriminate, and confusion matrix analysis proved that classification became more stable. The analysis of feature importance and Partial Dependence Plots offered readable information on clinically significant predictors. The suggested framework is more reliable, class-balanced, and interpretable, which justifies its use in decision support in automated fetal health monitoring systems.