A Variable Association Modeling and Dynamic Optimization Approach for Complex System Prediction

Abstract

In this paper, a two-stage modeling framework integrating multivariate statistical analysis and dynamic correction is proposed to address the problems of lack of variable correlation and insufficient modeling of nonlinear influence mechanism in traditional prediction models in complex systems. The key explanatory variables are first screened out based on feature engineering, and the multiple linear regression prediction model and logistic regression model are constructed respectively. To further address the interference of unobserved variables on the prediction results, the prediction model of important influencing factors is innovatively established, and the prediction results are optimized by quantifying the interference intensity of external factors. The empirical study shows that the improved combination model achieves a large improvement in prediction accuracy compared with the traditional prediction model. The modeling method provides a new technical path for accurate prediction of complex systems by establishing an interpret-able variable association system and a dynamic correction mechanism.