Mobile Phone Application Data for Activity Plan Generation

TL;DR

This paper introduces a novel generative model that combines mobile phone application data with travel surveys to produce realistic activity-travel plans for a large population, overcoming data limitations and enhancing activity-based transport modeling.

Contribution

The paper presents a new model integrating mobile app data with travel surveys, addressing data sparsity and bias, and demonstrating its effectiveness in generating activity schedules for over 263,000 individuals.

Findings

Generated plans are comparable to large-scale agent-based models.

Model significantly outperforms dummy models using only mobile data.

Approach is adaptable to other regions and data sources.

Abstract

Activity-based models in transport are crucial for providing a comprehensive and realistic understanding of individuals' activity-travel patterns. Traditionally, travel surveys have been used to develop these models, but they are often costly and have small sample sizes. Mobile phone application data, one example of emerging data sources, offers an alternative with wider population coverage over extended periods for developing activity-based models. However, the challenges of using these data include sampling biases in the population coverage and individual-level data sparsity due to intermittent and irregular data collection. To synthesise activity-travel plans, we propose a novel model that combines mobile phone application data with travel survey data, addressing their limitations. Our generative model simulates multiple average weekday activity schedules for over 263,000 individuals living in Sweden, approximately 2.6% of Sweden's population. We also introduce a temporal-score approach to improve home and work location identification approaches. We assess the model's performance against an existing large-scale agent-based model of Sweden (SySMo) and a dummy model using only mobile application data. The generated activity-travel plans are comparable to the SySMo model's output and significantly surpass the dummy model's results, suggesting the proposed model's capability to generate reasonable activity-travel schedules. The proposed model is adaptable to other regions with similar travel surveys and emerging data sources, like call detail records, advancing the use of these data for activity-based models in a cost-effective, easily updated manner.