Advanced Voting Method for Improving Random Forest Classification Algorithm Performance in Machine Learning

The Random Forest Classi f icat ion Algorithm is a popular Ensemble learning algori thm which deals wi th c lass i f ica tion of da ta with given set of a t t ribu tes on the basis of majori ty vo tes f rom various decision trees o f that fores t (Bre iman,Cut ler,2004). Classi f ica tion on the basis of majori ty votes by the decision t rees i s not be best way to pred ict c lass i f ica tion s ince di f ferent decision t rees may have dif ferent level of accuracy a t making pred ict ions. We conducted an experiment on d if ferent dataset wi th varying number of at t ribu tes and the accuracy fo r class i f ica tion of data on the basis of majori ty vo tes vs. weighted vot ing(on the basis of per formance of each tree in the fores t on the basis of i t s F1 score) was compared. I t was found that the weighted vot ing on the basis of F1 score ou tperforms the m ajori ty voting method in class i f ica tion accuracy. Here in th is paper we propose the vo ting mechanism in Random Forest Classi f icat ion for solving classi f ication problem should be on the basis o f F1 scores o f the decision t rees.