Machine Learning-Based Optimization of Timing and Intelligent Determination of Fetal Abnormalities in Non-Invasive Prenatal Testing

Abstract

Non-Invasive Prenatal Testing (NIPT) is a crucial technique in modern prenatal screening, holding significant value in detecting fetal chromosomal abnormalities. However, accurately predicting fetal Y chromosome concentration and optimizing the testing timing remain challenges in current clinical practice. This study employs machine learning methods to analyze the relationship between maternal physiological indicators and fetal Y chromosome concentration, aiming to construct a predictive model for optimizing NIPT testing strategies. The research begins with rigorous data quality control and preprocessing, including handling missing values, correcting outliers, and standardizing data. Through Pearson correlation analysis and coefficient of variation selection, gestational age, maternal BMI, and age were identified as core features. Addressing the severe multicollinearity among features, the study innovatively proposes a Gradient Boosting Random Forest hybrid model, effectively combining the anti-multicollinearity advantage of Random Forest with the sequential optimization capability of Gradient Boosting. Model validation results demonstrate that the hybrid model has significant predictive ability. Feature importance analysis further verifies the core roles of gestational age, BMI, and age in prediction. This research provides a new method for the intelligent prediction of fetal Y chromosome concentration in NIPT, contributing positively to improving detection accuracy and supporting clinical decision-making.