Investigating Mobility Patterns through Crowd-sourced Activity Data

Abstract

Abstract Objectives Investigating trip purposes represents an important phase of travel demand modeling which allows to correctly infer mobility patterns and to better understand travel behavior. Until now, researchers collected information on the motivation for performing a trip mainly through travel surveys. However, traditional methods of acquiring this type of information are challenging and expensive to implement; therefore, they are typically performed infrequently and with low sampling rates. These two occurrences do not always allow for adequate representation of the heterogeneity of trip purposes. This paper aims to investigate trip purposes through a novel approach that combines GPS-based data, such as Floating Car Data (FCD), and aggregated activity data available through open-source platforms, such as Google Popular Times (GPT), to better understand travel behavior. Material and Methods This research employs clustering techniques to categorize FCD into Home-Work trips and Not Home-Work trips. The latter category is further examined based on arrival times and stopover durations. This exploration utilizes activity patterns derived from GPT data, encompassing daily visit distribution and average visit duration obtained from user-shared mobile phone geo-traces. Results The methodology has been applied to a FCD dataset containing trips carried out between September and November 2020 in the EUR district of Rome, Italy. Through our approach, we generate 96 Origin-Destination matrices for Home-Work and Not Home-Work trips. By analyzing GPT data, 6 distinct activity patterns are identified within the study area, which allows for further segmentation of the Not Home-Work matrices. Conclusions This research presents an innovative method of inferring trip purposes for travel demand modeling. Exploiting the integration of FCD and GPT data, it enhances the representation of trip heterogeneity and the understanding of mobility patterns compared to traditional survey methods. Although challenges remain in handling purpose assignments for specific clusters, the comparison of computed metrics with existing literature results validates the approach reliability and aligns with expected behavior. Clinical Relevance Not applicable