一种降低误分类代价的权值分布优化算法

Optimized Weight Updating Method for Reducing the Cumulative Misclassification Cost

提出一种降低误分类代价的权值分布优化算法,其核心思想是在连续的提升过程中获得一系列的弱分类器,根据训练实例代价因子的大小及分类的结果不断更新训练集的分布,以减小累积的误分类代价。验证明这种改进后的算法减小了累积误分类代价。

This paper presents a optimized method to reduce the cumulative misclassification cost. The method uses the cost of misclassification to update the training distribution on successive boosting rounds. The experimental result shows that the optimized method can reduce misclassification cost.