Fetal hypoxia is a condition that necessitates timely and accurate diagnosis to mitigate the risks associated with fetal mortality. The analysis of FHR signals is a reliable method for identifying hypoxic conditions. However, challenges such as class imbalance in datasets and the lack of interpretability in ML models limit their adoption in clinical settings. This study proposes an approach to address these challenges by integrating advanced data balancing techniques, ML models, and explainability tools. The methodology employs the XGBoost classifier for FHR signal-based classification, leveraging three types of feature extraction: morphological, time-domain, and frequency-domain features. Various data balancing techniques, including SMOTE, ADASYN, GAN, and SMOTE-Tomek, were systematically evaluated. SMOTE emerged as the most effective method, achieving a better accuracy of 86.45% and an F1-score of 84.47%. To ensure clinical relevance, the study incorporated SHAP decision plots to enhance the interpretability of the model. The findings of this research underscore the importance of integrating feature-rich data, advanced data balancing techniques, XGBoost algorithms, and explainability to improve the diagnostic accuracy and clinical applicability of fetal hypoxia detection systems.