Network Classification with Missing Information

Abstract

Demand for effective methods of analyzing networks has emerged with the growth of accessible data, particularly for incomplete networks. Even as means for data collection advance, incomplete information remains a reality for numerous reasons. Data can be obscured by excessive noise. Surveys for information typically contain some non-respondents. In other cases, simple inaccessibility restricts observation. Also, for illicit groups, we are confronted with attempts to conceal important elements or their propagation of false information. In the real-world, it is difficult to determine when the observed network is both accurate and complete. In this paper, we consider a method for classification of incomplete networks. We classify real-world networks into technological, social, information, and biological categories by their structural features using supervised learning techniques. In contrast to the current method of training models with only complete information, we examine the effects of training our classification model with both complete and incomplete network information. This technique enables our model to learn how to recognize and classify other incomplete networks. The representation of incomplete networks at various stages of completeness allows the machine to examine the nuances of incomplete networks. By allowing the machine to study incomplete networks, its ability to recognize and classify other incomplete networks improves drastically. Our method requires minimal computational effort and can accomplish an efficient classification. The results strongly confirm the effectiveness of training a classification model with incomplete network information.