关键词: Logistic模型, 非平衡数据, ROC曲线, AUC值, 平衡化的五折交叉验证

Abstract: Based on the UCI database, this paper analyzes the sensitivity of the logistic model to different degree of unbalanced data. The research shows that: (1) the higher the degree of unbalanced data is, the poorer the ability that logistic regression identifying the rare classes is. (2)Compared to other corrected methods, OSS method is neither significant nor stable; Simple sampling has better performance relative to complex sampling.(3) AUC is not suitable for model selection under the condition of unbalanced data. Because it cannot distinguish the four corrected methods effectively, and cannot tell the differences before and after correction.

Key words: Logistic Model, Unbalanced Data, ROC Curve, AUC, Balanced 5-CV