TangentBoost is a robust boosting algorithm. The method combines loss function and weak classifiers. In addition, TangentBoost gives penalties not only misclassification but also true classification margin in order to get more stable classifiers. Despite the fact that the method is good one in object tracking, propensity scores are obtained improperly in the algorithm. The problem causes mislabeling of observations in the statistical classification. In this paper, there is a correction proposal for TangentBoost algorithm. After the correction on the algorithm, there is a simulation study for the new algorithm. The results show that correction on the algorithm is useful for binary classification.