Comparison between traditional model and causal intervention model: A case study baesd on directed acyclic graph
Huaqing Jiang, Xiaohui Yang · 2021 3rd International Conference on Machine Learning, Big Data and Business Intelligence (MLBDBI) · 2021
With the development of machine learning, the advantages of causal learning over statistics have been pointed out. This paper investigates whether causal logic can significantly improve machine learning performance by comparing the performance of the traditional model and the model after causal intervention. However, there is no unified process and lack of real data application in this kind of research now. To this end, the paper proposed a process of exploring hidden variables, constructing directed acyclic graphs, and eliminating confounding through backdoor criteria, to explore the cause-and-effect relationship. At the same time, this paper applies the strategy to establish the causal intervention strategies under the multiple linear regression model to forecast the grade of human living environment. Finally, the results of grade classification according to the specified criteria were compared with the results of traditional multiple linear regression model without causal intervention strategy, and it was found that the model with causal intervention strategy had better performance, which may suggest that causal learning is likely to eliminate or reduce the negative effects of pseudo-correlation in traditional models.