Refined Weighted Random Forest and Its Application to Credit Card Fraud Detection.

Random forest (RF) is widely used in many applications due to good classification performance. However, its voting mechanism assumes that all base classifiers have the same weight. In fact, it is more reasonable that some have relatively high weights while some have relatively low weights because the randomization of bootstrap sampling and attributes selecting cannot guarantee all trees have the same ability of making decision. We mainly focus on the weighted voting mechanism and then propose a novel weighted RF in this paper. Experiments on 6 public datasets illustrate that our method outperforms the RF and another weighted RF. We apply our method to credit card fraud detection and experiments also show that our method is the best.