Identification of Human Mobility Modes using Socio-Spatio-Temporal Predictive Models
The scientific analysis of the regularities observed in individual and collective human movement trajectories is of fundamental relevance to a wide range of areas urban planning, the prevention of epidemics, and natural security issues such as detection of clandestine activity, to name but a few. The ubiquity of mobile phones and location-based social media has enabled the capture of comprehensive, time-resolved individual information, offering a unique opportunity to observe human activity on an unprecedented scale. Indeed, recent theoretical developments suggest that a perfect algorithm can predict a person’s whereabouts with almost 90% certainty, given past observations of their location visits. Yet, major gaps remain in our understanding of human mobility dynamics.
In particular, (i) any practical predictive algorithm far underperforms the theoretical limit. This is primarily due to analyzing movement trajectories from a pure frequentist perspective, capturing a limited portion of the mechanisms behind human dynamics. (ii) Furthermore, the frequentist approach can at best capture averaged predictability, and says little about temporal predictability, which is of far greater relevance to national security issues. In this proposal, we will develop a comprehensive, data-driven framework that addresses the aforementioned issues. Our methodology will rest on two foundations: (i) First, we will go beyond static predictability measures and define a dynamic predictability by leveraging the multidimensional features that influence an individualÕs movement. Primarily, these are a combination of their social contacts (S), their spatial location (L) and the time of the day (T), or their socio-spatial- temporal context. (ii) In combination, the socio-spatio-temporal aspects form a multilevel network that has a clear structure, which we term activity modes, a previously undiscovered facet of mobility.
From these building blocks, we will develop a practical algorithm that will bridge the sizable gap between theoretical and usable predictability, make temporal predictions of human movement, and identify anomalous patterns. We will pay particular attention to scenarios of critical importance to national security. Finally, we will implement a series of data collection campaigns, using it to validate our methods and claims.
The proposed research will evolve along the following thrusts: Thrust 1: Defining and inferring mobility context: Will focus on a novel multidimensional quantitative characterization of the human trajectories which we term mobility contexts. Thrust 2: Mobility structure through contextual modes Will leverage mobility contexts to identify functional and structural clusters (activity modes) characterizing coarse-grained motifs of movement. Thrust 3: Anomaly detection and classification of travelers Will build a method capable of identifying anomalous changes in movement behavior, distinguishing between ordinary changes in visitation preferences from structural changes as measured by the activity modes. Thrust 4: Validation and optimization through high-resolution data will validate the methodology on a comprehensive series of datasets, relevant to national security and defense.
