Thresholding a Random Forest Classifier

摘要: The original Random Forest derives the final result with respect to number of leaf nodes voted for corresponding class. Each node is treated equally and class most votes wins. Certain in topology have better classification accuracies others often lead a wrong decision. Also performance forest different classes differs due uneven proportions. In this work, novel voting mechanism introduced: each has an individual weight. decision not determined by majority but rather linear combination weights leading more robust This method inspired construction strong classifier using small rules thumb (AdaBoost). Small fluctuations which are caused use binary trees balanced. Experimental results on several datasets object recognition action demonstrate that our successfully improves accuracy algorithm.