This is the authors' abstract. We don't add key points for this paper.
Clinical risk assessment in fetal health monitoring via cardiotocography (CTG) is frequently compromised by significant class imbalance and physiological outliers. These factors often degrade the sensitivity of machine learning models to critical pathological states. To address these challenges, this study proposes a synergistic integrated pipeline designated as PRO-LGBM (Purified, Rebalanced, and Optimized LightGBM), designed to sequentially address sensor-induced noise, minority class underrepresentation, and the complexities of high-dimensional hyperparameter landscapes. Specifically, the pipeline integrates Isolation Forest for outlier purification, Borderline-SMOTE for targeted class rebalancing, and Bayesian Optimization with the Tree-structured Parzen Estimator (TPE) for precision tuning of the Light Gradient Boosting Machine (LightGBM). Experimental results, validated through stratified 10-fold cross-validation and an independent hold-out test set, demonstrate that the proposed configuration achieves a peak accuracy of 96.61%. This performance significantly outperforms existing baseline models and state-of-the-art benchmarks. Statistical validation using the Wilcoxon Signed-Rank Test confirmed the significance of these performance gains with a p -value of 0.0020. Additionally, 95% Confidence Intervals substantiated the predictive stability of the model. Furthermore, SHAP-based interpretability provides a transparent diagnostic layer. It identifies abnormal short-term variability and prolonged decelerations as the primary clinical indicators of fetal distress. Computational analysis further confirms that the proposed synergistic pipeline maintains quasilinear scalability and negligible inference latency ( < 50 ms) to ensure real-time utility. By bridging advanced machine learning with clinical transparency, this study establishes a robust and trustworthy foundation for real-world obstetric decision-support systems.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Ilham et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: