FETAL HEALTH RISK STATUS IDENTIFICATION SYSTEM BASED ON CARDIOTOCOGRAPHY DATA USING EXTREME GRADIENT BOOSTING WITH ISOLATION FOREST AS OUTLIER DETECTION

Premature birth and birth defects contribute significantly to infant mortality, highlighting the need for early identification of fetal health risks. This study uses XGBoost for fetal health classification, integrating IForest for outlier detection to improve model performance. By varying the contamination percentage, learning rate (η), maximum depth, and n_estimator, the best results were achieved at CP = 8%, η = 0.01, max_depth = 7, and n_estimator = 100, which resulted in 100% accuracy, sensitivity, and specificity with a calculation time of 0.36 seconds. IForest effectively reduced the dataset from 2126 to 1956 samples by removing outliers, improving accuracy by 3.76%, and reducing computation time by 0.51 seconds. These findings suggest that IForest improves classification efficiency while maintaining high predictive performance, supporting early identification of fetal health risks to aid timely medical intervention.