Douglas Do Couto Teixeira

Given the difficulties in predicting human behavior, one may wish to establish bounds on our ability to accurately perform such predictions. In the case of mobility-related behavior, there exists a fundamental technique to estimate the predictability of an individual’s mobility, as expressed in a given dataset. Although useful in several scenarios, this technique focused on human mobility as a monolithic entity, which poses challenges to understanding different types of behavior that may be hard to predict. In this paper, we propose to study predictability in terms of two components of human mobility: routine and novelty, where routine is related to preferential returns, and novelty is related to exploration. Viewing one’s mobility in terms of these two components allows us to identify important patterns about the predictability of one’s mobility.Additionally, we argue that mobility behavior in the novelty component is hard to predict if we rely on the history of visited locations (as the predictability technique does), and therefore we here focus on analyzing what affects the predictability of one’s routine. To that end, we propose a technique that allows us to (i) quantify the effect of novelty on predictability, and (ii) gauge how much one’s routine deviates from a reference routine that is completely predictable, therefore estimating the amount of hard-to-predict behavior in one’s routine. Finally, we rely on previously proposed metrics, as well as a newly proposed one, to understand what affects the predictability of a person’s routine. Our experiments show that our metrics are able to capture most of the variability in one’s routine (adjusted $R^{2}$ R 2 of up to 84.9% and 96.0% on a GPS and CDR datasets, respectively), and that routine behavior can be largely explained by three types of patterns: (i) stationary patterns, in which a person stays in her current location for a given time period, (ii) regular visits, in which people visit a few preferred locations with occasional visits to other places, and (iii) diversity of trajectories, in which people change the order in which they visit certain locations.

Predicting mobility-related behavior is an important yet challenging task. On the one hand, factors such as one’s routine or preferences for a few favorite locations may help in predicting their mobility. On the other hand, several contextual factors, such as variations in individual preferences, weather, traffic, or even a person’s social contacts, can affect mobility patterns and make its modeling significantly more challenging. A fundamental approach to study mobility-related behavior is to assess how predictable such behavior is, deriving theoretical limits on the accuracy that a prediction model can achieve given a specific dataset. This approach focuses on the inherent nature and fundamental patterns of human behavior captured in that dataset, filtering out factors that depend on the specificities of the prediction method adopted. However, the current state-of-the-art method to estimate predictability in human mobility suffers from two major limitations: low interpretability and hardness to incorporate external factors that are known to help mobility prediction (i.e., contextual information). In this article, we revisit this state-of-the-art method, aiming at tackling these limitations. Specifically, we conduct a thorough analysis of how this widely used method works by looking into two different metrics that are easier to understand and, at the same time, capture reasonably well the effects of the original technique. We evaluate these metrics in the context of two different mobility prediction tasks, notably, next cell and next distinct cell prediction, which have different degrees of difficulty. Additionally, we propose alternative strategies to incorporate different types of contextual information into the existing technique. Our evaluation of these strategies offer quantitative measures of the impact of adding context to the predictability estimate, revealing the challenges associated with doing so in practical scenarios.

Free-Floating Car-Sharing (FFCS) services are a flexible alternative to car ownership. These transportation services show highly dynamic usage both over different hours of the day, and across different city areas. In this work, we study the problem of predicting FFCS demand patterns—a problem of great importance to the adequate provisioning of the service. We tackle both the prediction of the demand (i) over time and (ii) over space. We rely on months of real FFCS rides in Vancouver, which constitute our ground truth. We enrich this data with detailed socio-demographic information obtained from large open-data repositories to predict usage patterns. Our aim is to offer a thorough comparison of several machine-learning algorithms in terms of accuracy and ease of training, and to assess the effectiveness of current state-of-the-art approaches to address the prediction problem. Our results show that it is possible to predict the future usage with relative errors down to 10%, while the spatial prediction can be estimated with relative errors of about 40%. Our study also uncovers the socio-demographic features that most strongly correlate with FFCS usage, providing interesting insights for providers interested in offering services in new regions.

Free Floating Car Sharing (FFCS) services are a flexible alternative to car ownership. These transportation services show highly dynamic usage both over different hours of the day, and across different city areas. In this work, we study the problem of predicting FFCS demand patterns -- a problem of great importance to an adequate provisioning of the service. We tackle both the prediction of the demand i) over time and ii) over space. We rely on months of real FFCS rides in Vancouver, which constitute our ground truth. We enrich this data with detailed socio-demographic information obtained from large open-data repositories to predict usage patterns. Our aim is to offer a thorough comparison of several machine learning algorithms in terms of accuracy and easiness of training, and to assess the effectiveness of current state-of-art approaches to address the prediction problem. Our results show that it is possible to predict the future usage with relative errors down to 10%, and the spatial prediction can be estimated with relative errors of about 40%. Our study also uncovered the socio-demographic features that most strongly correlate with FFCS usage, providing interesting insights for providers opening service in new regions.