不完全数据集上的半参组群差异检测

Semi-EL detection of group differences between incomplete datasets

组群差异检测广泛应用于医药、社会网络等领域.现有的组群差异检测都是建立在数据集没有缺失的情况下,因此讨论并给出了在数据集缺失的情况下进行组群差异检测的方法.首先,使用一种新颖的缺失数据填充方法填充缺失数据,然后在半参环境下使用经验似然方法对得到的完全数据集估计出置信区间,进而进行组群差异检测.

Group difference detection had widely been used in medical research and social network analysis. It was studied a group-interaction approach for incomplete data mining. Specifically, the incomplete data were first completed by missing value imputation in the parimputation strategy, and then semi-empirical likelihood （semi-EL） inference was used to estimate the group differences. To experimentally illustrate the efficiency, the proposed approach was evaluated with UCI datasets.