An ensemble variable selection method based on conditional mutual information

With the advent of the era of big data, the data we come into contact with are gradually developing from low dimension to higher dimension. In order to make data analysis effective and efficient, variable selection is gradually rising. This paper introduces an ensemble variable selection method CMIENS based on conditional mutual information(CMI). The reasons for advocating that are as follows: Firstly, if and only if two random variables are conditionally independent, CMI is zero, which indicates that it can detect nonlinear conditional correlation. Secondly, compared with others, it has a better ability to control the type I error rates and a larger empirical power, which can well capture the relationship between variables. Finally, its calculation is simple and easy to implement. The results of numerical simulation experiments and real cases show that compared with other variable selection methods based on conditional correlation coefficient, CMI-ENS can not only select the correct variables better, but also has great advantages in running time.