A framework for using mobile phone data in population synthesis for agent-based modeling

In this study, we introduce a methodology that follows the conventional approach of generating a synthetic population (SynPop), which is subsequently enhanced and refined by geographical information using mobile phone data (MPD). The overarching goal is to enable a more specific trip distribution based on the established 4-stage model of transport demand modeling in our study area the city of Darmstadt and its surrounding rural areas in Germany. By employing a mixed exact-probabilistic data-matching approach, the methodology enhances representativeness and granularity, addressing biases and limitations of current datasets. Overall, six key attributes with a mixed approach of exact and non-exact matching were utilized for the matching process. The matching was executed iteratively using a greedy algorithm, continuing until no data points remained. A comparison of the two applied distance metrics, Nearest Neighbor matching using Propensity Scores and Mahalanobis distance each with and without a caliper, revealed, after validating the balance, that the Mahalanobis distance without a caliper achieved more reliable matching for our data. Nevertheless, the data basis must be more precisely aligned with the matching parameters in the future.