Developed countries have invested considerable eforts in promoting sustainable transportation modes, identifying bicycle use as a more environmentally friendly alternative. Nevertheless, this transition necessitates a deeper understanding of bicycle user behaviour to facilitate the process. This study investigates bike users' route selection behaviour based on real-world data. Data collected over six months from Bologna, Italy, were analysed to understand the flow and choices of bike users in a medium-small historical city. We employed a tessellation algorithm combined with Open Street Map street tags to examine bike users' preferences regarding urban trajectories. The dataset comprises of 290,117 unique trips consisting of 60,414,481 Global Positioning System (GPS) points. The results indicate that bicycles are predominantly used for short- to medium-distance trips. Specifically, in suburban regions, bike users tend to use larger roads leading to the city centre, whereas, in the city centre, they prefer routes that bypass the narrow streets of the historic area. These findings ofer insights into route choice factors and can guide improvements to cycling infrastructure for a safer, more eficient urban environment.
In response to climate change, energy crises, fossil fuel
depletion, overpopulation, and trafic congestion, sustainable
living has garnered increased attention. In densely
populated urban areas, promoting sustainable transportation is
vital. Encouraging cycling can reduce the use of motorised
vehicles and decrease trafic volumes. Cycling is a
sustainable, healthy, and economical transport mode, particularly
suitable for short-distance trips in historic and narrow urban
spaces. Facilitating bicycle use for commuting and leisure
as a sustainable transport option requires a thorough
understanding of urban mobility and travel behaviour. The role of
data mining in mobility research and urban cycling studies is
crucial. To assist in this cause, mobility studies can uncover
patterns in daily and routine human movements [
Although various models for analysing bike users’ behaviour have been proposed, a detailed investigation of route selection between two locations and multiple alternative trajectories is still lacking. Moreover, despite the precision ofered by GPS data, it is crucial to preserve users’ privacy. This paper addresses this gap by utilising mobility data from the Bella Mossa initiative, provided by the Public Transport Authority SRM Reti e Mobilità Srl, and the Open Street Map (OSM) data to examine cycling patterns in Bologna, a historic university city with narrow streets in northern Italy. We conducted an analysis of bicycle usage based on 290,117 self-reported unique trips, consisting of 60,414,481 GPS points, spanning six months, from April 2017 to September 2017. Specifically, we applied tessellation and clustering techniques to analyse mobility flows in urban areas based on origin-destination tiles and to study the trajectory selections of bike users.
This study aims to investigate how cycling infrastructure influences bike users’ route choices by utilising OSM street-type tags while preserving users’ privacy. To achieve this, the exact routes from GPS tracks are transformed into coarser shifts between origin and destination, retaining useful information for analysis. The following are the highlights of this study. Firstly, we investigate the utilisation of various types of infrastructure, including urban highways, diferent bike lane configurations, and streets, to understand bike users’ preferences regarding route choice. Secondly, we identify the most frequently used flows and routes for transportation purposes in Bologna, shedding light on preferred routes. Lastly, we examine how road types and their characteristics influence bike users’ route choices, providing insights into the factors influencing route selection.
The obtained results reveal that cyclists tend to favour streets with lower trafic volumes and dedicated cycling infrastructure. Furthermore, cyclists prefer to avoid narrow streets in the old town when possible. Understanding these patterns helps identify opportunities to improve infrastructure and create a safer urban environment. For instance, identifying the most frequently used origin-destination tiles and the preferred paths between them can help guide infrastructure development and promotion. Also, the findings will guide recommendations to promote cycling as a viable urban transportation option, and the proposed approach, despite the conclusions being specific for the dataset, the pipeline can be applied in similar urban contexts.
Cycling ofers a sustainable, eco-friendly, cost-efective, and
inclusive solution for urban mobility [
The pervasive collection of tracking data has enabled
extensive trajectory data collection for applications like
intelligent transport systems, urban management, and
environmental protection [
Trajectory analysis is essential for understanding route
selection in urban environments. This process involves
choosing a specific path from an origin to a destination based
on various characteristics [
In this paper, we propose a method based on tessellation and flow analysis algorithms to study the decision-making process of urban bike users. Specifically, we leveraged information from OSM and addressed privacy concerns related to the use of GPS data through a clustering algorithm.
The Bella Mossa1 data was collected from the municipal area of Bologna, a historical city in northern Italy. The Bella Mossa initiative was a 6-month program to promote a healthy lifestyle and sustainable mobility from 01/04/2017 to 30/09/2017. Around 15,000 unique participants recorded 895,000 trips for a total of 3.7 million travelled kilometres. Participants earned redeemable points for cycling, walking, and using public transport by using a GPS-enabled smartphone app that recorded trips and anonymously sent the data to a database. Each dataset entry includes a unique activity ID, a user-selected activity type descriptor, and details about the timestamp, GPS position, and accuracy. The activity ID enables trip reconstruction without being traceable to the participant. In this study, we specifically focus on data points labelled as cycling activities. The dataset is anonymised; individual user data has not been analysed, and only aggregated analysis results are presented.
We conducted pre-processing to clean and correct data, addressing incomplete or inaccurate values. Preliminary analysis identified extremely short or long activities. Short activities with fewer than two GPS points were removed. Long activities with pauses longer than six minutes or where the bike user stayed in one place for over six minutes were split into multiple segments. Activities longer than two hours or over 30km after segmenting were excluded, as they likely included multiple segments or nonurban activities (e.g., sports). Activities under 100m or less than three minutes were also removed. Finally, activities with GPS points outside Bologna or timestamps outside the study period were excluded. The initial dataset of 320,109 unique activities with 72,396,179 GPS points was reduced 1Due to Italian legislation restrictions, we are unable to publicly release the datasets. Any requests can be submitted to SRM Reti e Mobilità Srl. to 290,117 unique activities consisting of 60,414,481 GPS points, recorded every 3-10 seconds.
In this section, we outline the proposed method using the Bella Mossa dataset and the OSM street network data. First, we employed a map-matching algorithm to refine the bike users’ trajectories, converting sequences of GPS points into sequences of road segments. Next, we applied tessellation and clustering algorithms for trajectory and road type analysis, while preserving users’ privacy.
The map-matching algorithm assigns original GPS
coordinates to the most plausible road network segments,
improving the accuracy of GPS data and linking these coordinates
to the corresponding road network information. In practice,
we used the Open-Source Routing Machine (OSRM) project
and the Hidden Markov Model (HMM) approach with the
Viterbi path-finding algorithm [
We used a spatial tessellation algorithm to subdivide the
city area into tiles according to the GPS points densities
[
Combining the representation of users’ trips with road segments –coarser than GPS points– and map tessellation – which identifies areas rather than precise points– makes it possible to protect users’ privacy. Moreover, as shown in Figure 2, the irregular tessellation enables a more focused analysis of the city’s most frequently used areas by increasing the density of the tiles.
Flow analysis was utilised to explore the decision-making process of bike users regarding urban trajectories. We examined cyclists’ trajectories from an origin Voronoi tile to a designated destination one, with the origin and destination tiles determined using the initial and final GPS points. This approach resulted in a comprehensive table summarising all movements between all Voronoi tile pairs. Then detailed trajectory analysis was conducted on the top twenty most frequent flows, merging flows from opposite directions to simplify distance calculation and grouping.
Given the complexity of the road network, multiple routes
exist between origin-destination pairs. To identify the most
common paths, we performed a trajectory clustering
analysis using the Fréchet distance [
The clustering process divided all trajectories within a flow into distinct clusters based on the routes taken, with cluster sizes ranging from 1 to over 200 trajectories. Clusters containing fewer than 5 trajectories were deemed less significant for the final analysis. Finally, leveraging the OSM street tags, the road type analysis allows us to explore the road infrastructure primarily used by cyclists. For computational eficiency, the road type analysis focused on data for the month of April, while the entire dataset was utilised for lfow analysis. We specifically computed the road type usage proportion by dividing the lengths of all road types in the activity by the total activity length. Similarly, we extended the road type analysis to the trajectory clusters to ascertain whether street type influences bike users’ route selection.
In this section, we briefly present the results discussing the main outcome of the route choices and road type analysis.
Highlight 1: Road Network Usage - Figure 1 depicts the result of the map-matching process. Notably, the cleaning procedure enables the extraction of noise-free trajectories represented as sequences of road segments. This preprocessing activity is of the utmost importance to reifne subsequent analyses and obtain meaningful results. In particular, the origin GPS points of the cleaned trajectories were used to tessellate the city area into tiles according to the GPS points densities. The resulting irregular tessellation with Voronoi diagrams is shown in Figure 2. The tiles with the highest number of activity GPS points were located in the city centre.
Highlight 2: Frequent Flows - Flow analysis utilised tessellation tiles, assigning each activity an origin and destination tile. Activities with the same origin-destination pair were grouped and counted as one flow. We performed trajectory analysis on the 20 most frequent flows, employing trajectory clustering to identify diferent routes between locations. Figure 3 shows the most popular routes among the 20 top flows.
Highlight 3: Route Choices - Each flow was analysed to identify various routes between tile pairs. Trajectories were clustered into similar route groups using the DBSCAN algorithm, and a street-type analysis was conducted –based on the OSM road type labels– to assess its influence on route selection. For instance, Figure 4a illustrates the diferent
Due to environmental concerns and urbanisation, there has been a growing emphasis on alternative modes of transportation, such as cycling. This paper presents a framework for analysing mobility flows within urban settings, examining the route choices between pairs of origin-destination tiles. An in-depth analysis of the top 20 flows showed that route choices were notably afected by one-way streets. According to the OSM road type labels and the results, cyclists typically avoid travelling against the designated direction, even when permitted. Furthermore, streets with cycling infrastructure are preferred, while narrow ones are generally avoided.
For future work, we plan to compare the real trajectories with potential shortest and quasi-shortest paths to better understand why cyclists choose specific routes. Additionally, we intend to investigate whether the efectiveness of the clustering algorithm may impact the final inference. Also, we plan to investigate the generalisability of the proposed framework to other transportation modes, exploring diferent tessellation strategies or clustering techniques to extract the sub-network of each mode.