A Quantitative Association Model for Male Fetal Y Chromosome Concentration Based on Random Forest Regression and SHAP Interpretability

Abstract

This study aims to clarify the quantitative association patterns between maternal biological indicators and fetal Y chromosome concentration in male fetuses. First, comprehensive preprocessing was performed on 549 male fetal detection datasets, including key missing value deletion, Z-score outlier handling, and construction of derived features such as BMI grouping and gestational age segmentation, to ensure data quality and applicability for modeling. Spearman's rank correlation analysis confirmed a significant monotonically negative relationship between maternal BMI and Y chromosome concentration. Subsequently, a random forest regression model was established to capture the nonlinear relationship between features and Y concentration. Model parameters were optimized via grid search and 5-fold cross-validation, ultimately determining the number of decision trees T=800. On the test set, the model demonstrated excellent performance with a coefficient of determination R² and a mean absolute error MAE of 0.0183 ng/mL. Further SHAP interpretability analysis quantified feature contributions, revealing that X chromosome concentration, number of blood draws, and Y chromosome Z-score were the top three factors influencing Y concentration prediction. Additionally, SHAP dependency plots clearly demonstrated a sharp increase in the positive contribution of gestational age during weeks 12–15, while high BMI systematically reduced Y concentration.