关键词: 多标签分类, 准确性评估, 不平衡类

Abstract: Classification is one of the hotspots of data mining.The multi-label classifier can predict a set of labels for one data object.Data objects which have same attributes but their corresponding labels are different，and the number of them is imbalanced.The existing evaluation algorithm can not distinguish the cost.A weights-based accuracy evaluation method EMOWDIF（Evaluation Method Based on Weights Difference） is proposed.The weight is calculated according to the ratio of the multi-label instances，giving different rewards to the classifier，and can distinguish the performance of different classifiers effectively.It is programmed and used to evaluate the classification result.Experiments show the mothod can get a better performance on evaluating the classifier.

Key words: multi label classify, accuracy evaluation, class-imbalance