Machine Learning-driven Causal Inference Methods: Progress, Evaluation and Multi-Domain Applications

Ever since R.A. Fisher proposed the Randomized Controlled Trial (RCT), causal inference has been formalized as a rigorous theoretical science. Causal inference has now shown potential and good results in many fields. Researchers are applying it to more complex and diverse domain scenarios. This has led to the progression of causal inference from statistical methods to machine learning, deep learning, and representation learning. Many different schools of thought have been born in this process, employing different algorithms to enhance the processing power and accuracy of causal inference methods. They all have their own theoretical merits and their own appropriate scenario areas. Therefore, this study hopes to systematically sort out these methods along the lines of timeline, method schools, etc., and establish an evaluation framework based on scenario-priority indicators, and then give an algorithmic decision matrix based on this framework to form an evaluation and decision system between scenario-indicator-algorithm. And in the end, a classical dataset is selected to validate the proposed decision matrix, which is hoped to facilitate the researchers.