Mocha: a framework to characterize and compare mobility

Abstract

The evaluation of mobile network algorithms via simulation is strongly dependent of the accurate representation of the mobility scenario. i.e., the mobility models and/or traces used in simulation should reflect reality as much as possible. There are numerous traces in the literature that are commonly used as benchmark to validate mobility models, generators and wireless networking solutions. Thus, considering the heterogeneityof the real mobility datasets that are publicly available, it is crucial, if not necessary, to understand the differences and similarities among them and also their characteristics. In this work we propose MOCHA, a MObility CHaracterization Framework that extract, classify and compare mobility properties, based on their marginal statistical distribution. These properties cover different aspects of human mobility and can bedivided into three categories: social, spatial and temporal. MOCHA is composed by three different modules: a parser, a property extractor and a classifier. The parser is responsible for converting any mobility dataset into a standard form that can be understood by MOCHA. The extractor is responsible for extracting up to 11 mobility properties from the parsed dataset. Finally, the classifier analyzes the marginal statistical distribution of each mobility property, splits them into head and tails, and classifythem according to the tails. Once that MOCHA extract and classify each property, it is possible to visualize the similarities among different datasets by using a k-means algorithm and the Principal Component Analysis method. We validate our proposed framework with 13 real mobility traces that are frequently used as benchmark, in additionto 18 synthetic traces generated by different and well known mobility models. By comparing real and synthetic datasets, our results show how the considered mobility generation tools fail to reproduce realistic scenarios. We also show how MOCHAs methodology is more elegant than their predecessors and more complete, for considering all the steps from the original trace parsing process to the mobility propertiesclassification.