Elevating Maternal Healthcare: Synergy of Cardiotocography, Machine Learning Models and Interpretive Analysis

The global goal of reducing maternal mortality and improving neonatal health, in particular during labor where clinical actions must be precise and timely, is important. Abnormalities such as low or high maternal oxygen levels, inappropriate uterine contraction patterns, or abnormal fetal heart rates during delivery can endanger the well-being of the mother and child. Currently, many essential signs are primarily assessed and recorded using visual means, which are susceptible to delay or even error. However, of late, machine learning has been characterized as a cutting-edge approach that helps improve practice in clinical settings by generating more accurate forecasts based on intricate data patterns This study’s objective is to formulate an appropriate machine-learning model for predicting maternal and fetal health risks at delivery for improved clinical outcomes. Within the freely accessible cardiotocography dataset consisting of the number of uterine contractions more than and the fetal heart rate of 21 features, Recursive feature elimination (RFE) folds the features down into 15. To solve the problem of the class imbalance MSMOTE (Modified Synthetic Minority Over-sampling Technique) is used for this purpose. A number of the machine learning procedures such as Random Forest, LightGBM, CatBoost, AdaBoost, Decision Tree, K-Nearest Neighbor, Support Vector Machine, Logistic Regression, Multi-Layer Perceptron, and Gradient Boosting amongst others are also trained and subjected to evaluation. Of these, CatBoost and LightGBM models recorded the highest level of accuracy at 96.0% and 95.1% respectively. To further interpret as well as validate the model predictions, LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations) used.