Revisiting pitfalls of DTN datasets statistical analysis

Abstract

Contact traces collected in real situations represent a popular material to assess the performance of a Delay Tolerant Network. These traces usually require some preprocessing to be fully usable. Especially, several assumptions can be made prior to performing the statistical analysis of contact and inter-contact times. We first classify these assumptions, and analyze their impact on the statistical characterization of three well-known datasets. We also identify some pitfalls in dataset analysis that might strongly influence the conclusion made by the experimenter. Based on our own experience, we subsequently propose a preliminary checklist to help researchers avoid undesired ambiguities or misunderstandings in further studies.