Comparison of Tree-based Feature Selection Algorithms on Biological Omics Dataset
Zheng Liu, Jiayuan Song
Abstract
Feature selection is of great significance in processing high-dimensional data, which can save cost of computation and improve the performance of analysis. Tree based classifiers have been gaining their popularity due to their great performance and their extended feature selection methods also have been widely adopted for dimension reduction of high dimensional dataset. However, which specific tree-based feature selection method is most suitable for feature selection task for omics dataset has not been comprehensive investigated. In this work, we compare the performance of different tree-based feature selection (SVM, Random Forest, Logistic Regression) methods on high-dimensional data of biological omics. The results indicate that GBDT performs best, XGBoost performs slightly worse than GBDT while the performance of RF is the worst.