Adaboost algorithm with floating threshold

A novel AdaBoost algorithm with floating threshold, called AdaBoost.FT, was put forward based on the maximum likelihood principle. The proposed AdaBoost.FT algorithm significantly improved the stability of classification compared to the real AdaBoost algorithm. For this purpose, each weak classifier of AdaBoost.FT algorithm used the floating threshold to obtain the outputs of classifiers by the distribution on the training samples. In contrast, the real AdaBoost algorithm employing the fixed classification threshold was so unstable that the classified results were oversensitive to the slight change of the instance near to the classification threshold. Furthermore, the using method about AdaBoost.FT algorithm was elaborated. Theoretical analysis and experimental results both show that AdaBoost.FT algorithm was effective.