A1185
Title: Stacked ensemble modeling for pregnancy health monitoring data with measurement error
Authors: Jia-Ren Tsai - Fu Jen Catholic University (Taiwan) [presenting]
Abstract: The purpose is to establish an ensemble prediction framework to construct a regression model between repeated measurements of systolic and diastolic blood pressure and serum creatinine concentration in pregnant women. Given the issues of measurement error in practical data, multiple estimation methods that account for measurement errors are integrated using a stacking ensemble technique to build a robust fused parameter vector for enhanced predictive accuracy. Two fusion strategies are proposed in this study. The first computes a weighted average of model parameters based on performance metrics such as mean squared error. The second builds a secondary regression model, using predictions from the base models as covariates, to learn optimal weights. Simulation results demonstrate that the proposed ensemble methods maintain stable performance across varying sample sizes and reliability ratios. In a real application involving repeated measurements of blood pressure and serum creatinine in pregnant women, the ensemble models outperform individual models in terms of both bias and prediction error. These findings highlight the effectiveness of combining multiple measurement error correction techniques with model averaging and stacking to enhance prediction accuracy in the presence of measurement error.
