Dealing with incompleteness in multidimensional analysis of health records: An experience on fetal growth

Abstract

Relational and multidimensional datasets are often affected by incompleteness. To cope with this problem, several strategies have been proposed in literature, often depending on the incompleteness type and on the specific application domain. Majority of approaches draw hints from the data already available in the same database in order to fill up missing values, but this can be unsuitable when dealing with legitimate missing data, dynamic scenarios and anonymized data, which are very common for example in medical databases. To deal with these kinds of incompleteness, we propose a new approach to provide indicators about the statistical relevance of the analyzed data. A prototype based on a specific modeling strategy and on binary data structures, has been implemented to test the feasibility and the effectiveness of the proposed approach on a real dataset about fetal growth.