A Generalized K2 Algorithm for Learning Bayesian Network Structures Using Ridge Regression

In this article, we propose a new approach of the K2 algorithm to enhance the process of selecting candidate parents in Bayesian networks by incorporating dependency criteria based on ridge regression. Traditional K2 algorithms primarily rely on scoring functions to evaluate parent sets, which may not effectively capture all dependencies in the presence of multicollinearity. Our approach leverages ridge regression to address this limitation by penalizing overly complex models, thereby providing a more robust mechanism for parent selection. Through real-world data applications, we demonstrate that our generalized algorithm significantly improves the accuracy and reliability of parent selection in Bayesian networks. The findings suggest that this integration of ridge regression with the K2 algorithm offers a promising avenue for advancing Bayesian network structure learning, particularly in complex data.