Optimizing Fetal Health Predictions with Machine Learning Models Using Cardiotocography Data

Our project mainly works on classifying fetal health conditions using machine learning techniques applied to Cardiotocography (CTG) data, which includes various fetal and maternal health features such as baseline values, accelerations, decelerations, and variability measures. Comprehensive data preprocessing steps were undertaken, including handling missing values, removing duplicates, normalizing continuous variables, and removing outliers using interquartile ranges to ensure data quality. To address class imbalances, the Synthetic Minority Oversampling Technique (SMOTE) was employed. Several ml models, including SVM, NB, KNN, RF, DT, and LR were trained to predict fetal health as Normal, Suspect, or Pathologic, with performance evaluated using metrics and confusion matrices. To enhance prediction reliability and robustness, an ensemble learning approach was implemented, utilizing a majority voting method to aggregate predictions from LR, DT, and RF models. This approach leverages the individual strengths of these models while reducing biases, demonstrating the effectiveness of ensemble techniques in improving classification performance and ensuring more reliable predictions for fetal health conditions.