Predicting encounter and colocation events in metropolitan areas

TL;DR

This paper introduces a novel Bayesian model that predicts encounter and colocation events in urban areas by leveraging spatio-temporal regularities, improving accuracy over existing methods on large WiFi and cellular datasets.

Contribution

The paper presents a new encounter and colocation predictive model that exploits spatio-temporal regularity, outperforming standard Bayesian and state-of-the-art predictors.

Findings

Improved prediction accuracy over naive Bayesian methods.

Effective on large-scale WiFi and cellular datasets.

Demonstrates the importance of spatio-temporal features in mobility prediction.

Abstract

Despite an extensive literature has been devoted to mine and model mobility features, forecasting where, when and whom people will encounter/colocate still deserve further research efforts. Forecasting people's encounter and colocation features is the key point for the success of many applications ranging from epidemiology to the design of new networking paradigms and services such as delay tolerant and opportunistic networks. While many algorithms which rely on both mobility and social information have been proposed, we propose a novel encounter and colocation predictive model which predicts user's encounter and colocation events and their features by exploiting the spatio-temporal regularity in the history of these events. We adopt weighted features Bayesian predictor and evaluate its accuracy on two large scales WiFi and cellular datasets. Results show that our approach could improve prediction accuracy w.r.t standard naive Bayesian and some of the state-of-the-art predictors.