关键词: 不平衡数据, 分类, 迁移学习, Precision-Recall曲线

Abstract: Traditional classification algorithms based on the balance data meet some challenges, when data distribution become more and more imbalanced. Transfer learning can solve the problem of imbalanced data distribution by using the relevant auxiliary data sets to compensate the imbalanced target data set. In this paper, we proposed the UnbalancedTrAdaboost(UBTA) binary classification algorithm based on TrAdaboost, which calculates the weights of weak classifiers usingthe auprc (the Area Under the Precision-Recall Curve) of different classes and updates the weights of misclassified data of different classes with different mechanisms. The AUC measure is more accurate combined with G-mean and BER when evaluated the unbalanced classification, since AUC is insensitive to changes in class distribution. The results of these three metrics indicate that, the UBTA algorithm achieves better performance for imbalanced data and classifies more minority instances with the high accuracy of majority instances.

Key words: Imbalanced Data, Classification, Transfer Learning, Precision-Recall Curve