Abstract: It constructs a Feature Weighted MPM（FWMPM） based on the Boosting. It estimates the relative importance of each feature by computing the Boosted distance. It is also the weight of each feature. It makes use of the weights for computing the inner product and Euclidean distance in kernel functions. In this way the MPM can allay the influence of trivial relevant feature. The experimental results and analysis show that the FWMPM has the better performance of class than the standard MPM.

Key words: Minimax Probability Machine（MPM）, Boosting, information gain, feature weighting

摘要： 在最小最大概率机中引入Boosting权值确定方法，构造特征加权最小最大概率机（FWMPM）。利用Boosting方法计算各个特征对分类任务的重要度，把此特征重要度作为原始数据各个特征的权重，对核函数中的内积和欧氏距离进行加权计算，从而可以减轻最小最大概率机被一些弱相关的特征影响。实验结果和理论分析表明，该方法比标准最小最大概率机具有更好的分类性能。

关键词: 最小最大概率机, Boosting, 信息增益, 特征加权