Abstract
Human mobility knowledge is key for urban planning or mobility models design. Therefore, estimating reliable mobility parameters is crucial to lay an unbiased foundation. However, most works estimating such features rely on datasets made up of the history of mobile network cells where the user is located when she makes active use of the network, known as Call Data Records (CDRs), or every time the her device connects to a new cell, without taking into account cell changes not caused by movement. Could we accurately characterize human mobility with such datasets? In this work we consider three approaches to collect network-based mobility data, propose three filtering techniques to delete cell changes not caused by movement and compare mobility features extracted from the traces collected with each approach. The analysis unveils the need for a filtering step to avoid important biases, and the negative impact that using CDRs may have in estimating mobility parameters.