关键词: 因果发现, 因果推断, 因果关系, 机器学习

Abstract: Causal discovery techniques have broad applications in fields such as medicine, biology, economics, and social sciences, enabling the identification of causal relationships between variables, thereby enhancing the interpretability of prediction results. Causal discovery refers to the process of identifying causal relationships among variables from data, including determining whether a causal relationship exists between variables and understanding the direction and strength of such relationships. Currently, causal discovery faces challenges such as effectively leveraging high-dimensional data, precisely controlling confounding variables, and managing complex interactions among variables. Traditional causal discovery methods are based on conditional independence tests, heavily relying on data quality and performing poorly in high-dimensional contexts. Machine learning techniques have significantly advanced the development of causal discovery, including efficient data processing and analysis, uncertainty estimation, and credibility analysis. This review summarizes the current progress in causal discovery： it introduces representative methods in traditional causal discovery, exploring the issues in their core processes; subsequently, it summarizes popular causal discovery methods in the field of statistical learning, detailing their core ideas and comparing their performance and applicable data types and scenarios. The primary objective is to provide more valuable references for researchers in the fields of data science and statistical learning. Finally, future research directions for causal discovery are summarized.

Key words: causal discovery, causal inference, causal relationship, machine learning