Setting of class weights in random forest for small-sample data
Gao Zhi-kun · Computer Engineering and Applications Journal · 2009
Random forest has been proved to be an efficient algorithm for classification and feature selection in bioinformatics.Although the effect of parameter setting on results is very limited,a group of appropriate parameters can generate excellent performance.This paper focuses on the setting of class weights in random forest to deal with classification and feature selection problems of unbalanced small-sample data and determines the optimal class weight.In order to compare the performance of feature selection with different weights,SVM is applied in the paper.The results show that optimal class weight is variable and cannot form a standard.However,people can find some weights with which not only classification but also feature selection can get better performance.