Transforming GPS tracks for planning and services

Truly understanding how a city moves means going beyond the simple sum of journeys.
A recent study conducted on these analyses highlights a key step: using large volumes of GPS data to feed activity-based demand models, thereby shifting the focus from “how people travel” to “why they travel.”
This change in perspective is useful for those who govern and make decisions about urban mobility, because it allows for an understanding of daily routines and decision chains that a traditional trip-based model tends to break down, considering trips as simple and isolated journeys that take you from point A to point B.

From GPS pings to mobility diaries. Building a reliable database

For decades, surveys on travel using questionnaires have been the basis for planning.
They remain valuable, but they have clear limitations: high costs, long timeframes, and distortions due to respondents’ memories, known as recall bias.
Research shows that GPS data, collected by providers and anonymized, offer an advantage in terms of scale and accuracy.

One example is the study conducted on the metropolitan city of Milan, which involved 100 million pings and almost one million users observed for one month, between mid-November and mid-December 2024.
This data requires thorough cleaning.
The work being done aims to remove profiles with few scattered pings and then refine it further, resulting in an 82% reduction in the sample compared to the initial data and a final core of 17,000 reconstructable mobility diaries.

The key methodological step is to distinguish, within a chaotic sequence of GPS points, between movement and stops, i.e., activity.
The fixed time or speed thresholds typical of traditional methods do not work well when pings are irregular. In this case, added value can be found in the DBSCAN program, an innovative clustering algorithm.
What the program does is apply the algorithm to coordinates weighted by the speed between one point and the next.
In the pattern that is thus created, the faster and more dispersed travel points end up in clusters, while the stopping points, with almost zero speed and greater density, are classified as noise.

The insight is to reverse the meaning: the insight is to reverse the meaning: here, the anomaly goes from being considered an error to becoming a sign of activity.
To increase the robustness of the data collected, further corrections and post-processing are then applied, such as the K Nearest Neighbors technique.

Segmentation, scenarios, and testing of mobility policies
To understand the reasons behind these movements, we need to interpret the destination.
The classification takes place on two levels.
The first level has three macro categories: Home, Main Activity, and Secondary Activity, defined by time and frequency parameters. The second level has a detailed plan obtained by integrating data from land use.
The stops are overlaid by cross-referencing them with information from cartographic databases and information on the territory (OpenStreetMap).

What is thus defined is a catchment area capable of covering 98% of the activities identified and processing approximately 1.2 million buildings and 224,000 points of interest.
In this way, a long stop in a university area can be interpreted as Study, while a stop in an office area can be interpreted as Work.
However, there remains an important point for decision-makers: land use maps can have gaps and ambiguities, so the initial macro classification is crucial to avoid misleading interpretations.

From here, we move on to user segmentation, moving beyond the idea of average travelers. By analyzing recurring patterns, profiles emerge such as full-time workers, part-time workers, university or secondary school students, and a group of non-workers or retirees, with explicit limitations due to the absence of personal data.
However, this approach allows us to reconstruct distinct behaviors on a large scale, without recall bias.

The value that can be added to public administration lies in the use that can be made of this data. These methodologies can be included in the planning toolbox and feed simulation platforms capable of building a digital twin of the population under observation.
This makes it more realistic to test how cities would react to new subway lines, redesigned services, and incentives, or to optimize public transport schedules and plan cycling infrastructure in a targeted manner.
The condition implicit in all this work is to treat data as infrastructure: quality, cleanliness, integration, and readability become part of the decision, not an incidental technical step.

Sources:
https://datamobility.it/magazine/decodificare-il-ritmo-delle-citta-con-i-dati-gps/

Ben-Akiva, M.E., Bowman, J.L. (1998). Activity Based Travel Demand Model Systems. In: Marcotte, P., Nguyen, S. (eds) Equilibrium and Advanced Transportation Modelling. Centre for Research on Transportation. Springer, Boston, MA.
https://doi.org/10.1007/978-1-4615-5757-9_2