Section: New Results

Deciphering Predictability Limits in Human Mobility

Participants : Douglas Do Couto Teixiera, Aline Carneiro Viana, Jussara Almeida [Federal U. of Minas Gerais] , Mario S. Alvim [Federal U. of Minas Gerais] .

Human mobility has been studied from different perspectives. One approach addresses predictability, deriving theoretical limits on the accuracy that any prediction model can achieve in a given dataset. Measuring the predictability of any phenomenon is a very useful, but hard task, and especially so in the case of human behavior. Such complexity is due to the uncertain and heterogeneous behavior of humans, as well as to the variability of parameters influencing such behavior. Predictability is concerned with the maximum theoretical accuracy that an ideal prediction model could achieve in a scenario expressed by a given dataset. As such, unlike particular comparisons of alternative prediction models on different datasets, it does not depend on a specific prediction strategy but rather on human behavior, as captured by the available data. Besides, it does not rely on the tuning of a multitude of sensible parameters, providing instead a parameter-free view of how predictable human mobility can be (as expressed in the data).

This approach focuses on the inherent nature and fundamental patterns of human behavior captured in the dataset, filtering out factors that depend on the specificities of the prediction method adopted. In this work, we revisit the state-of-the-art method for estimating the predictability of a person's mobility, which, despite being widely adopted, suffers from low interpretability and disregards external factors that have been suggested to improve predictability estimation, notably the use of contextual information (e.g., weather, day of the week, and time of the day). We propose a new measure, regularity, which together with stationarity, helps us understand what makes a person's mobility trajectory more or less predictable, as captured by Song et al.'s technique. We show that these two simple measures are complementary and jointly are able to explain most of the variation in Song et al.'s predictability. As such, we here use them as proxies of that technique to analyze how one's mobility predictability varies.

Additionally, we investigate strategies to incorporate different types of contextual information into predictability estimates. In particular, we were the first to quantify the impact of different types of contextual information on predictability in human mobility, for different prediction tasks and datasets. Our results show that, for the next place prediction problem, the use of contextual information plays a larger role than one's history of visited locations in estimating their predictability. Finally, we propose and evaluate alternative estimates of predictability which, while being much easier to interpret, provide comparable results to the state-of-the-art. We show that these estimators, while being more interpretable, provide comparable results in terms of predictability.

This paper was published at ACM SIGSPATIAL 2019, a A+-ranked conference in our domain, and was indicated as a top-six best paper candidate. An extended version is being prepared for submission to a journal.