Precise Estimation of Fetal Y Chromosome Concentration Based on XGBoost Regression and Entropy-Weighted TOPSIS Optimization

Abstract

Addressing challenges in fetal Y chromosome concentration estimation for non-invasive prenatal testing (NIPT)—including inconsistent measurement units, significant noise, weak monotonic correlations among variables, and strong potential nonlinearity—this study focuses on constructing and optimizing a Y chromosome concentration prediction model. First, bubble plots and correlation analysis revealed significant weak nonlinear relationships between fetal Y chromosome concentration and gestational age at testing, as well as maternal BMI. Building upon this, an XGBoost regression model was developed as the primary predictor and compared with Support Vector Machine Regression. Results demonstrated significantly improved model fit for XGBoost, achieving value of 0.958 and a low standardized root mean square error of 0.707, substantially outperforming the SVR model. To further enhance model robustness and interpretability, this study introduced the entropy weighting method to determine objective weights for key features. Combined with the Two-Operator Partial Indexed System of Selection method, this approach performed weighted optimization of model parameters. The optimized model successfully simplified complex nonlinear relationships into a clearer structure while significantly improving fitting accuracy for extreme and marginal samples. Ultimately, paired t-tests on the same sample error confirmed that the improvement of the optimized model over the baseline model was highly statistically significant.