부분최소자승 회귀분석을 이용한 변수 간 인과관계탐색 알고리즘

다국어 초록 (Multilingual Abstract)

In recent years, there have been many attempts to develop and improve causal discovery algorithms (CDA) for learning structures of Bayesian networks. Constraint-based CDAs implement conditional independence (CI) tests to verify dependency between variables so as to find a directed acyclic graph which is observationally equivalent to the noninterventional data. Multiple Search (MS) algorithm exploits multiple linear regression for CI test in which multiple search is possible in a CI test so as to improve both speed and accuracy. However, the CI test based on multiple linear regression might be unreliable if there exists multicollinearity among variables. To deal with the multicollinearity problem, we propose a robust CI test using partial least squares (PLS). Using PLS regression, the proposed method performs better when applied to data with multicollinearity. We show the robustness of our method by a numerical example.

목차 (Table of Contents)

• Abstract
• 1. Introduction
• 2. Background
• 3. Proposed algorithm
• 4. Numerical example
• 5. Conclusion
• 6. References