Abstract: Researchers used to compare the typical version with enhanced one proposed by themselves through the results after certain cycles.To verify and complement this method，the author researched on the general character of classifying results，defining three quantitative indicators—error rate in steady state，regulating scale and oscillation，separately corresponding to accuracy，convergence speed and stability，thus constructing a relatively comprehensive evaluation system.

Key words: AdaBoost, weak classifier, error rate, weighted parameter

摘要： 工程实际中，往往通过对比两个AdaBoost算法在相同弱分类器数量条件下的错分率来比较算法性能，这样就忽略了在弱分类器数量增加时，错分率的波动会造成对比不准确的问题。为此，分别针对分类器性能的分类准确率、收敛速度和稳定性，提出了稳态错分率、调节规模、振荡度三个量化指标，构成了一个相对完备的评价体系。实验表明，该评价体系能更全面反映AdaBoost的分类效果。

关键词: AdaBoost, 弱分类器, 错分率, 权重系数