Fetal Abnormality Detection: Exploring Trends Using Machine Learning and Explainable AI

Monitoring fetal health and classifying it into different classes is crucial to prevent adverse situations. Healthy fetus leads to healthy new generations. Early detection and treatment of any medical condition is supreme for the well-being and safety of both mother and newborn. This work presents a novel approach to fetal health classification using machine learning models. Three classes—Normal, Suspect, and Pathological—are created by selecting pertinent features from the Cardiotocography (CTG) dataset. Predictive models are created using machine learning (ML) techniques, such as XGBoost, Voting Classifier, Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), and Logistic Regression (LR). XGBoost model has performed well and attained an accuracy-94%. RF model gave the highest precision and recall for classifying normal and suspect classes. Furthermore, this work also uses Explainable AI (XAI) tool called SHAP (SHapley Additive exPlanations) to know/interpret how RF model produces the outcome for test data. ML embedded with XAI supports healthcare providers to trust the model as well as timely intervention of doctors leads to better pregnancy outcomes and maternal-infant health.