Prediction and interpretation of fetal health status in the womb with machine learning method using tocogram data

This study utilized the publicly available fetal health dataset containing cardiotocogram data to evaluate the performance of various machine learning algorithms for predicting the status of fetal health. The algorithms were implemented with Python and the PyCaret library, and Synthetic Minority Over-sampling Technique (called SMOTE) used for countering the imbalanced distribution of the target variable. Notably, the Light Gradient Boosting Machine exhibited the highest accuracy and F1 score, as well as the lowest severity of misclassifications according to our penalty score system that considered the differing severities of misclassification errors. The study demonstrated the potential of machine learning algorithms to accurately predict fetal health and enhance clinical decision making processes. Validation of the models on more diverse and larger datasets is recommended. The study also showcased the utility of user friendly libraries like PyCaret, highlighting how clinics could potentially build their own machine learning models with lesser effort.