Random forest ensemble using a weight-adjusted voting algorithm
Ahhyoun Kim, Jeeae Myung, Hyunjoong Kim · Journal of the Korean Data and Information Science Society · 2020
A classification ensemble is a learning method that aggregates different classifiers to obtain more accurate class predictions. Among many developed methods, Random Forest is known as one of the most accurate ensemble methods. It combines many randomized decision trees using simple majority voting scheme. Wave, a weighted voting algorithm, has proven to outperform simple majority voting when combined with bagging. In this paper, we investigated whether random forest using the wave voting scheme can further improve classification accuracy. Experiments show that the larger the ensemble size, the more accurate it is than other methods including single tree, bagging, AdaBoost, random forest (simple majority voting). The result also shows that random forest with wave is more accurate than bagging with wave when the ensemble size is large enough.