Evaluation of How Methods for Creating People Flow Data Affect Performance of Epidemic Routing

Abstract

Recently various datasets regarding human mobility have been available for research purpose. People Flow Project by Center for Spatial Information Science provides people flow data of Japanese urban areas created from Person Trip Census data. There exist some data in which the number of points of departure and arrival is only one for each representative postal address although real human mobility traces does not have this property. It comes from the reason that Person Trip Census does not include coordinates of longitude and latitude information, but does only representative postal addresses. When we consider performance evaluation of information dissemination by epidemic routing, effects of methods for creating the data on simulation results have not been well considered. In this study, we investigate this effects through creating three types of people flow data using Fukui’s Person Trip Census to compare the performance of epidemic routing. The first data is that the points of departure and destination are assigned only one for each representative postal address, and nodes expressing persons or cars move linearly between them. The second data is that the points of departure and destination are the same with the first one, but each node moves along a shortest-path route on the road network. The third data is that the points of departure and destination are assigned to randomly selected points on the road distributed around the addresses and each node moves along a shortest path route. We perform numerical simulations of epidemic routing where a selected node initially have a message to spread to other nodes within a communication distance. We evaluate the impact on temporal epidemic size by changing the communication distance. As a result, we find that when the distance is sufficiently short, the speed of dissemination in the third data becomes much slower than those of the first and second ones. This result indicates we must be careful to use people flow data for evaluating the performance because a illusory pattern that a large number of mobility traces gather into the representative points of departure and arrival cannot be negligible.