Class Imbalanced NIPT Data Processing based on Ensemble Learning: Focusing on Early Detection of Chromosomal Abnormalities in Female Fetuses

Abstract

To address class imbalance in non-invasive prenatal testing (NIPT) data for pregnant women, improve the accuracy of early prenatal detection for female fetuses, and reduce the risk of shortened treatment windows, this paper comprehensively compares the classification performance of four ensemble learning models-EasyEnsemble, RUSBoost, AdaBoost, and BalancedRandomForest-for chromosomal aneuploidy labeling. This aims to determine the optimal approach for handling imbalanced data. First, the dataset underwent preprocessing: irrelevant feature columns were removed, categorical data was converted using label encoding, and the dataset was split into training (80%) and testing (20%) sets. Next, the trained models were applied to the testing set to generate predictions, which were then compared against the ground truth labels. Finally, classification results were visualized using confusion matrices and other charts, while classification performance metrics such as accuracy, precision, and F1 score were calculated to compare the classification effectiveness of the four models from different perspectives. Experimental results indicate that the four models exhibit differences in classifying chromosomal aneuploidy labels. Overall, the RUSBoost model demonstrates optimal classification performance with an accuracy rate of 87.91% and good recall, showcasing strong robustness and generalization capabilities. This provides a practical solution for early detection of female fetal abnormalities in clinical settings.