关键词: 不平衡数据, 类别不平衡学习, AdaBoost, AsymBoost, 阈值

Abstract: Quantities of imbalanced datasets exist in real world and classical learning algorithms attempt to get high precision on the datasets while learning few concepts from the minority class.Research on class-imbalance learning is an important part of machine learning.As an improvement to AdaBoost，AsymBoost improves performance on classification of examples of minority class but decreases accuracy of examples of majority class.AsymBoost may still have the overfitting problem when weights of examples become too large.In this paper a novel modified AdaBoost algorithm is proposed by new reweighting and label-modifying approach to deal with imbalanced datasets and it has shown promising results when compared to AdaBoost and AsymBoost.

Key words: imbalanced datasets, class-imbalance learning, AdaBoost, AsymBoost, threshold