【摘要】 在模式识别、机器学习以及数据挖掘中,分类是一个基本而又重要的问题。虽有大量的分类器应运而生,但由于处理不完整数据的复杂性,它们大都是针对完整数据的。然而,由于各种原因,现实中的数据通常是不完整的。因此,对不完整数据分类器的研究具有重要意义。通过分析以往在分类过程中对不完整数据的处理方法,提出了一种不完整数据分类器:DBCI。在DBCI的训练过程中,将缺失值的频数按比例地分配到其它观测值的频数中。因此,不完整数据集所包含的信息可以得到充分利用。在12个标准的不完整数据集上的实验结果表明,与分类效果显著的不完整数据分类器RBC相比,DBCI具有更高的分类效率和更稳定的性能,并且它的分类准确率可以与RBC相媲美。更多还原

【Abstract】 Classification is an elementary and important problem in pattern recognition,machine learning and data mining.Though many classifiers have been proposed,most of them deal with complete data,which is due to the complexity of dealing with incomplete data.Yet actual data sets are often incomplete because of various kinds of reason.So the study of classifiers for incomplete data is of great significance.With the analysis of main methods of processing incomplete data for classification,a new classifier for incomplete data denoted as DBCI is presented.In the training process of DBCI,frequencies of missing values are distributed proportionally across frequencies of other observed values.So the information contained in incomplete datasets can be sufficiently utilized.Experiments are carried out on twelve benchmark incomplete data sets.Compared to the remarkable Robust Bayes Classifier (RBC) that is very effective for classifying incomplete data,DBCI is more efficient and more stable and its classification accuracy is comparable to that of RBC.更多还原