Modelos de previsão de mobilidade humana usando dados de fontes heterogêneas

Abstract

Understand the mobility of large groups of people can help in urban planning, containment of diseases or even disaster contingency plans. Mobile data and georeferenced applications are valuable sources for examining the mobility of large groups. In this sense, the literature has several models that seek to describe or predict the patterns of human mobility in a particular region over a period of time. Most of these models were assessed using a single source data, e.g. data of mobile phone calls or GPS data from georeferenced applications. Therefore, the robustness of these models to different data types, and especially the combination of data from multiple sources (heterogeneous data), is still unknown. In this context, this dissertation proposes two models to predict human mobility that were designed to explore both mobile data and georeferenced data applications (in an isolated or combined way). The first model, called MobDatU, seeks to predict the mobility of a person in a target area and in a given time window based on the popularity of each region of the target area and the transition probabilities of people between two different regions. The second model, the MobDatU-Contact, seeks to predict the mobility of a person considering the contact relationship between people. The two new models as wells as two state-of-the-art models, namely SMOOTH and Leap Graph, were evaluated considering various scenarios single data source and multiple data source. The experiments indicate that the MobDatU always produces results that are better than or at least comparable to the best baseline in all scenarios, unlike the previous models whose performance is much more sensitive to the type of data used. Moreover, the MobDatU-Contact has produced better results than the MobDatU in all evaluated scenarios, showing that the location of the contacts of a person can be useful in predict the human mobility.