Abstract: In many real world applications, the number of examples from different class is significantly different, which means the number of examples in major class is much larger than that of minor class.Therefore, learning from imbalanced data set has received much attention of machine learning and data mining community.Considering that most of previous research focus on binary class problem, this paper proposes a multi-class imbalance method based on HDDT ensemble.Empirical study shows that the method is effective for multi-class imbalance learning.