异常信息的智能分类算法研究

Incomplete data belief classification algorithm based on adaptive KNN imputation

摘要:

摘 要：针对现有的基于插补的不完整数据分类方法的性能较差并且难以表征由于缺失值引起的不确定性等问题，提出了一种基于自适应KNN插补的不完整数据信任分类算法（BAI)。对于某一个不完整样本，该方法首先根据找到的近邻类别信息得到单个或多个版本的估计样本，这样在保证插补的准确性的同时能够有效地表征由于缺失引起的不精确性，然后用分类器分类带有估计值的样本。最后，在证据推理框架下提出一种新的信任分类方法，将难以划分类别的样本分配到对应的复合类来描述由于缺失值引起的样本类别的不确定性，同时降低错误分类的风险。用UCI数据库的真实数据集来验证算法的有效性，实验结果表明该算法能够有效地处理不完整数据分类问题。

Abstract:

Abstract: In order to address the problems that the incomplete data classification algorithms based on imputation strategy have poor performances and cannot characterize the uncertainty caused by missing values, an incomplete data belief classification algorithm based on adaptive KNN imputation (BAI) is proposed. This method yield one or more versions of estimations for a specific incomplete sample, which can not only ensures the accuracy of estimation, but also capture its imprecision. Then, the basic classifier that can implement in complete data well is employed here to classify the edited data with estimations. A new belief classification approach is developed in this paper to assign the indistinguishable object to the proper meta-classes so as to characterize the uncertainty of classification caused by missing values and decrease the risk of misclassification. The real data sets in the UCI are employed to verify the effectiveness of BAI, and the result represents that BAI can effectively handle the problem of incomplete data classification.