Predicting Traveling Distances and Unveiling Mobility and Activity Patterns of Individuals from Multisource Data

Abstract

This work investigates whether the user-generated data from multiple sources, such as smart cards and social media, can be used to identify main mobility/activity patterns based solely on geo-tagged information. To perform such an analysis, automated models are developed to (1) retrieve user mobility patterns from historical, user-generated data logs, (2) categorize users based on the similarity of their observed mobility patterns, and (3) predict the travel distances of users for participating in future activities. For testing purposes, user-generated data sets from smart card logs and Twitter profiles collected between November 2013 and February 2015 in London are used. User-generated data from 200 smart card and 32 active Twitter users are collected and 6 main clusters are identified based on the mobility/activity pattern similarities of users. Results show that it is possible to integrate data logs from multiple sources to capture the main mobility/activity patterns observed in an area. Results also reveal that the accuracy of the predicted travel distance of one user’s trip can be significantly improved if the user’s previous activities are considered in the prediction process.