Gaussian Process Regression Model Based on Cross-Validation Weighted Bagging Algorithm and Its Applications

Abstract

As a non-parametric regression technique widely applied across various disciplines, Gaussian Process Regression (GPR) faces certain challenges in the selection of the kernel function. To address this issue, this paper innovatively adopts a model aggregation strategy to dynamically determine the kernel function, rather than selecting a single model, thereby enhancing the model's adaptability and predictive power. Specifically, on one hand, the proposed method integrates Gaussian processes with different kernel function representations using a bagging algorithm. On the other hand, considering the varying importance of each sub-model, this paper employs cross-validation for weighted processing to balance the overfitting and underfitting issues of the predictive model. Finally, through extensive simulation experiments and real data analysis, the results demonstrate that the proposed method significantly improves the predictive capability compared to some classic predictive models, possesses better generalization performance, and can be effectively applied in various application scenarios.