Variable selection using Mean Decrease Accuracy and Mean Decrease Gini based on Random Forest 论文

摘要

Variable selection is very important for interpretation and prediction, especially for high dimensional datasets. In this paper, a new method is proposed based on Random Forest (RF) to select variables using Mean Decrease Accuracy (MDA) and Mean Decrease Gini (MDG). We also use dichotomy method to screen variables, which is proved to perform very fast. Experiments on 10 microarray datasets show that the new method is proficient and robust. In addition, we compared the proposed method with other variable selection methods, and the results demonstrated that our proposed method is more robust and more powerful in both accuracy and CPU time.