【摘要】 大数据背景下,现实生活中存在大量的非平衡数据,不同类别样本数量不平衡,而且在个体错分成本或错分损失上也不平衡。对于数量上的不平衡,已有许多成功的算法,典型的方法是在学习的过程中动态地改变样本个体的权重;但是针对错分成本的算法很少,原因之一是这种损失在实际问题中很难获得。对于数据集中每一个个体都潜在地存在可能给机构带来的错分损失,提出一个基于投影距离的错分损失期望的函数,并将这个函数用于数据分类集成算法中。分类集成算法的迭代过程中,弱分类器的选择原则是使得正确分类个体的权重之和与损失期望之和的加权和取最大值的弱分类器。在UCI数据集上的实验结果表明,在保持传统集成算法分类性能的基础上,所提算法能较好地提高少数类的分类性能。更多还原

【Abstract】 In the context of big data,there is a large amount of unbalanced data in real life,the number of samples in different categories is unbalanced,and the individual misclassified cost is also unbalanced. For quantitative imbalance,there are many successful algorithms. The typical method is to dynamically change the weight of the sample individual during the learning process. But there are very few algorithms for misclassified cost. One reason is that such loss is difficult to obtain in practical problems. For each individual in the data set,there is a potential misclassified loss that may be brought to the organization. A function based on the projection distance of misclassified loss expectation was proposed,which was used in the data classification ensemble algorithm. In the iterative process of the classification ensemble algorithm,the weak classifier was selected which makes the weighted sum of the weights of the correctly classified individuals and the sum of the loss expectations take the maximum value. Experimental results on the UCI dataset show that while mataining the classificaiton performance of traditional ensemble algorithm, the proposed algorithm can improve the classification performance of the minority class better.更多还原