Smart Rideshare Matching: Feasibility of Utilizing Personalized

Preferences

Shaguna Gupta

1,2 a

, Shi Feng

and B. Brian Park

4 b

Link Lab, University of Virginia, Charlottesville, Virginia, U.S.A.

School of Computer Science and Statistics, Trinity College Dublin, College Green, Dublin, Ireland

Link Lab, Department of Civil and Environmental Engineering, University of Virginia, Charlottesville, Virginia, U.S.A.

Link Lab, Departments of Civil & Environmental Engineering and Systems & Information Engineering,

Uni .S.A.

guptas2@tcd.ie, {sf5jh, bp6v}@virginia.edu

Keywords:

Rideshare Matching, Big Data Analytics, Vehicular Telematics Data.

Abstract:

We investigated the feasibility of utilizing vehicular telematics data for ride-sharing matching. The main focus

was to use personalized preferences including home and workplace departure and arrival times. A case study

was conducted using the vehicular telematics commuting data between their home to the University of Virginia

(UVA) campus. Using data from April 2022, which captures vehicle trips, arrival times, and departure times

at UVA, this research analyzed vehicle trips over two weeks to identify individuals with similar commuting

schedule preferences. By clustering vehicles based on proximity and timing, we proposed a framework for

matching individuals who share similar arrival and departure schedule preferences and live in nearby locations,

thereby facilitating coordinated ride-sharing opportunities. The ﬁndings are presented through visualizations

illustrating ride-matching potential, particularly during peak commuting hours. The matching would offer

a convenient ride-sharing solution for UVA commuters while maintaining their commuting ﬂexibility. This

approach could also offer a sustainable transportation solution that enhances travel efﬁciency, lowers environ-

mental impact, and supports the broader adoption of ride-sharing within academic and urban settings. The

proposed framework provides a scalable model for systematic ride-sharing implementation and could guide

future research and policy development for sustainable campus mobility solutions.

1 INTRODUCTION

Current transport-share systems or carpooling typi-

cally rely on users to actively request or offer a ride

and to coordinate the time and pickup location. Ser-

vices such as Lyft and Uber have addressed this prob-

lem by using location to provide ride services that are

convenient and on-demand. The on-demand and con-

venience aspects of transportation might also be the

main reason behind using personal cars as they allow

to combine commutes with other activities (e.g., pick-

ing up kids to and from school, running errands, go-

ing to off-campus meetings, etc.). This convenience,

however, comes at a great personal and societal cost,

including trafﬁc congestion and parking demand. De-

spite various agencies’ incentives and discounts for

ride-sharing, this kind of service has not been widely

https://orcid.org/0000-0002-9361-3097

https://orcid.org/0000-0003-4597-6368

used in daily commuting for obvious reasons men-

tioned above as well as hassled coordination, schedul-

ing requirements, commitment, and having to actively

request or offer rides.

In this research, we conducted a case study us-

ing a university community to increase engagement

in ride-sharing by analyzing vehicular telematics data

made available by Wejo. The analyses mainly focused

on drivers’ commutes to the University of Virginia

(UVA) and the potential of matching based on their

arrival times to the University and departure times

to their work-to-home trips. The UVA is central-

ized around the Charlottesville area and faces chal-

lenges in managing trafﬁc ﬂow due to its limited park-

ing spots. As campus activities fully resumed in

2022, making a return to normalcy after the disrup-

tions caused by the COVID-19 pandemic, the ﬂow of

trafﬁc to and from the campus noticeably increased.

Because of the ever-increasing parking demands on

the campus, there is an opportunity to optimize trafﬁc

482

Gupta, S., Feng, S. and Park, B. B.

Smart Rideshare Matching: Feasibility of Utilizing Personalized Preferences.

DOI: 10.5220/0013291400003941

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 11th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2025), pages 482-489

ISBN: 978-989-758-745-0; ISSN: 2184-495X

ﬂow using ride-sharing initiatives.

Ride-sharing can be an efﬁcient mode of sustain-

able transport for work colleagues or people who live

in close proximity, especially in areas with limited

public transport, walking, or cycling options. Ride-

sharing or car-pooling can be beneﬁcial for organi-

zations and individuals in terms of reduced travel

costs, improved parking efﬁciency, decreased trafﬁc

congestion, and positive environmental impacts. Be-

sides, it can be a productive, greener, and more sus-

tainable choice for individuals and the wider com-

munity. This research presents an analysis of vehi-

cle trips to and from UVA over a two-week period in

April 2022, with the goal of identifying potential ride-

sharing opportunities based on personalized commut-

ing arrival/departure time preferences.

The key objectives of this analysis are to mainly

recommend ride-sharing suggestions based on timing

and proximity. By identifying individuals who arrive

at UVA at similar times and leave UVA at similar de-

parture times, the aim is to enable coordinated ride-

sharing opportunities that align with their schedules.

Furthermore, the study suggests ride-sharing options

for individuals commuting from and to nearby loca-

tions relative to each other to enhance convenience.

These criteria are used to group vehicles based on

their arrival and departure locations, both at UVA and

at their home locations before and after the UVA trips.

This research details the methodology used to

identify these ride-sharing opportunities, evaluates

the results obtained from the analysis, and presents

visualizations to support the ﬁndings. The insights

gained from this study could be instrumental in de-

veloping a systematic ride-sharing program that en-

hances the commuting experience for UVA students,

faculty, and staff. We expect that the approach used

in this research is applicable to many similarly sit-

uated communities and institutions, where optimiz-

ing transportation systems can signiﬁcantly lower en-

vironmental impact, and improve overall travel efﬁ-

ciency. This framework tailored to different settings

can serve as a versatile solution for fostering sus-

tainable and convenient commuting practices in ur-

ban and academic environments alike. The remain-

der of this paper is organized as follows. Section 2

summarized the status of existing literature related to

car sharing, followed by section 3 with data and ap-

proaches used to analyze the vehicular data. Section

4 presents feasibility evaluation and results, followed

by section 5 discussion, section 6 conclusions and

section 7 future work.

2 LITERATURE REVIEW

Interest in ride-sharing to address trafﬁc, parking, and

energy issues in cities, companies, and college cam-

puses has driven extensive research on optimizing

models and understanding inﬂuencing factors. Our

study reviews current research on carpooling models,

preferences, and system improvements.

Studies have proposed models and algorithms

aimed to enhance car-sharing systems to beneﬁt both

users and car-sharing companies. Focusing on the

user perspective, Narman et al. presented a model

that employs a two-layer matching system (Narman

et al., 2021), and Hussain et al. proposed a system

speciﬁcally designed to optimize car sharing frame-

work for employees in large organizations (Hussain

et al., 2022), while Masoud and Jayakrishnan in-

troduced a real-time algorithm to address the ride-

matching problem within a ﬂexible ride-sharing sys-

tem (Masoud and Jayakrishnan, 2017). In the two-

layer model that Narman et al. developed, the ﬁrst

layer matches riders based on personal characteris-

tics, such as safety, punctuality, and comfort. The sec-

ond layer limits wait times with personalized thresh-

olds. A machine learning-based recommendation sys-

tem achieved 90 percent accuracy in predicting rider

preferences, providing successful matches and trip

completions. Considering car sharing in large compa-

nies, Hussain et al. developed a framework that con-

siders factors like home location, target destination,

time windows, and personal behavior to optimize car-

pooling groups. The system updates schedules in real

time, offering ﬂexible carpooling solutions. The pro-

posed framework efﬁciently manages recurrent travel

demand, especially for company employees. The

ﬂexible system that Masoud et al. proposed allows

for dynamic, real-time matching and multi-hop rides,

considering users’ preferences and minimizing wait-

ing times. Their algorithm can also solve large-scale

ride-matching problems quickly, providing comfort to

riders through optimal routing and reducing the num-

ber of transfers. These studies developed innovative

models using algorithmic approaches to enhance user

experiences in car-sharing systems, fostering greater

participation and efﬁciency in car-sharing systems.

Some researchers have also concentrated on opti-

mizing road networks, particularly in relation to road

congestion and capacity. De Palma et al. focused

on the impact of dynamic congestion on carpool-

ing matching in their paper (de Palma et al., 2022).

The study considered scheduling preferences and dy-

namic congestion in its ride-sharing framework. Re-

sults showed optimal matching occurs when drivers

and passengers are sequenced by location, but differ-

Smart Rideshare Matching: Feasibility of Utilizing Personalized Preferences

483

ing arrival times complicate matching and may intro-

duce tardiness penalties. Considering users’ choice

of ride-sharing, destination, and path, Xingyuan Li

et al. presented a ride-sharing trip-assignment model

using a bi-level programming approach for optimiz-

ing the reserve capacity of road networks. The pro-

gramming was then formatted into a single-layer op-

timization problem (Li et al., 2024). They concluded

that subsidizing ride-sharing drivers can enhance road

capacity, rivaling the effects of road expansion with-

out ride-sharing. Incorporating dynamic congestion

and user preferences, these studies show that effec-

tive ride-sharing strategies can signiﬁcantly improve

road networks and reduce congestion.

Other researchers have examined the social as-

pects of carpooling. Limited participation in carpool-

ing can be attributed to speciﬁc strategies employed

by ride-hailing companies, as well as concerns ex-

pressed by passengers. Naumov and Keith focused

on the economic and environmental impacts of ride-

hailing (Naumov and Keith, 2023). The study found

most ride-hailing trips, especially pooled rides, are

unproﬁtable due to subsidies. Adjusting pricing by

widening the gap between individual and pooled ride

costs could increase revenue, reduce vehicle miles

traveled, and beneﬁt both companies and the urban

environment. Linchao Li et al. explored what inﬂu-

ences college students’ views on carpooling by imple-

menting a multinomial logit model based on survey-

based data (Li et al., 2023). They concluded that con-

cerns about safety and cost are key reasons carpooling

is not popular among students. Safety and cost are pri-

mary concerns, a reliable carpooling information plat-

form for students could therefore increase support for

carpooling. These studies provided various insights

into how ride-sharing systems can be modeled, opti-

mized, and inﬂuenced by users’ preferences.

Despite the valuable ﬁndings provided by the re-

viewed studies, there exists a notable gap in the re-

search: few studies have utilized real-world vehicle

trajectory data that speciﬁcally capture drivers’ depar-

ture and arrival time preferences in their daily com-

muting trips. Most existing ride-sharing models rely

on simulations or survey data, which may not fully

reﬂect the dynamic nature of actual commuting pat-

terns. In addition, the integration of vehicle telematics

data has been under-explored in the context of opti-

mizing ride-sharing on university campuses commut-

ing patterns. To address this gap, our study proposes

an innovative approach that leverages real-world ve-

hicular telematics data to analyze commuting patterns

at the University of Virginia (UVA), and the poten-

tial for matching based on their arrival and depar-

ture times at UVA, which generates private route op-

tions for users. This research allows for more accu-

rate identiﬁcation of potential ride-sharing opportuni-

ties based on the timing and availability of potential

vehicle trips on campus, which aims to reduce traf-

ﬁc, improve efﬁciency, and create a more sustainable

transportation system in the city of Charlottesville.

3 MATERIALS AND METHODS

As noted, our research focuses on identifying patterns

where individual commuters arriving at and departing

from a university campus at similar times or from sim-

ilar locations could beneﬁt from shared transporta-

tion. Leveraging trip data, including Trip IDs, Vehi-

cle IDs, ignition status, and geographical coordinates,

we can cluster vehicles based on their arrival and de-

parture times and locations. This allows us to pro-

pose ride-sharing opportunities that could reduce traf-

ﬁc congestion, lower transportation costs, and con-

tribute to a more sustainable campus environment.

3.1 Vehicular Telematics Data

This research used vehicular telematics data made

available to the research team by the Virginia De-

partment of Transportation (VDOT). It is noted that

VDOT purchased the data from Wejo company. Out

of 26 weeks of data, this research utilized vehicle

trips for ten working days in April 2022, focusing

on Trip IDs, Vehicle IDs, ignition status, and geo-

graphical coordinates (latitude and longitude) to iden-

tify potential ride-sharing clusters. These ten consec-

utive weekdays were selected as other months were

affected by COVID-19 (some still teleworking during

Fall 2021) and many University academic activities

(e.g., Spring recess in March, graduation in May).

3.2 UVA Boundary

To accurately determine which trips were associated

with UVA, a boundary polygon was created to repre-

sent the UVA campus, including its parking lots (Uni-

versity of Virginia, 2023). This polygon was used to

ﬁlter trips that either started or ended within the UVA

boundary. Only trips that had their ignition turned

off (KEY OFF) within this boundary were consid-

ered as arriving at UVA, while trips that had their

ignition turned on (KEY ON) within the boundary

were considered as departures. Figure 1 shows the

UVA campus boundary on OpenStreetMap (Open-

StreetMap contributors, 2017) used to ﬁlter relevant

trips and this is being utilized to understand the ar-

rivals and departures in the subsequent sections.

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

484

Figure 1: UVA Polygon Boundary.

3.3 Vehicle Operational Schedule

Understanding

The initial phase in the methodology involved loading

and transforming the data. Timestamps recorded in

Coordinated Universal Time (UTC) were converted

to Eastern Time (EDT) to align with the local time

zone at UVA. The data was then ﬁltered to include

only relevant trips, speciﬁcally those between 5 AM

and 8 PM on weekdays, and within a predeﬁned set of

valid postal codes within the 25-mile radius of UVA.

3.3.1 Arrival and Departure Schedule

Based on the UVA polygon boundary, the ﬁltered data

was further processed to identify arrivals and depar-

tures at UVA. The study speciﬁcally analyzed trips

that ended or started within the UVA polygon area,

considering these as the ﬁrst trip of the day to UVA

and the last trip of the day from UVA. Based on this,

we ﬁnalized the trips that arrived or departed.

• Arrivals Extraction Algorithm

This algorithm is designed to ﬁnd the ﬁrst trip

each vehicle makes to UVA on any given day.

A trip is classiﬁed as an arrival at UVA when

the vehicle’s ignition is turned off, represented by

the KEY OFF ignition status. The algorithm pro-

cesses data related to vehicle trips and extracts the

earliest instance of a vehicle arriving at UVA on

each day within the dataset. This approach ﬁlters

the dataset to include only trips where the vehi-

cle’s ignition was turned off (i.e., arrival events)

and then sorts the data by vehicle and timestamp.

• Departures Extraction Algorithm

This algorithm identiﬁes the last trip for each ve-

hicle that departs from UVA on a given day. The

trip is considered a departure if the vehicle’s ig-

nition is turned on, represented by the KEY ON

ignition status. The algorithm identiﬁes the last

trip departing from UVA for each vehicle on that

day. This is designed to primarily extract the ﬁrst

Algorithm 1: First Trip Arrival Extraction Algorithm.

1: Input: DataFrame d f , Polygon uva polygon

2: Output: DataFrame f irst rows o f f irst tr ip

3: Convert Timestamp est to Trip Date (date only).

4: Filter rows where Ignition Status is KEY OFF.

5: Sort DataFrame by Vehicle ID and Timestamp est.

6: Group by Vehicle ID and Trip Date to get the ﬁrst

Trip ID of the day.

7: Join the dataset with the ﬁrst Trip ID to get the full

details of the ﬁrst trip.

8: Convert d f to Pandas format for spatial ﬁltering.

9: Convert back to Polars DataFrame.

10: Group by Vehicle ID and Trip Date to keep only the

ﬁrst rows of the ﬁrst trip.

11: Drop unnecessary columns.

12: Return the cleaned DataFrame.

Algorithm 2: Last Trip Departure Extraction Algorithm.

1: Input: DataFrame d f , Polygon uva polygon

2: Output: DataFrame f irst rows o f last tr ip

3: Create a new column Trip Date by extracting the date

from Timestamp est.

4: Filter d f where Ignition Status is KEY ON.

5: Sort d f by Vehicle ID and Timestamp est.

6: Group d f by Vehicle ID and Trip Date.

7: For each group, select the last Trip ID.

8: Join d f with last trip ids d f to get the last trip’s rows.

9: Convert df to Pandas for spatial ﬁltering.

10: Convert the ﬁltered DataFrame back to Polars.

11: Extract the last trip’s ﬁrst row for each vehicle and day.

12: Drop unnecessary columns.

13: Return the resulting DataFrame.

row of the last recorded trip for vehicle departures

from a deﬁned location based on its timestamp.

Both algorithms identify the ﬁrst trip and last trip

of each vehicle arriving at and departing from

UVA on a given day. The key steps involve ﬁl-

tering for relevant trips (ignition off/on), sorting,

grouping by vehicle and date, extracting the ﬁrst

trip, and cleaning the dataset to remove unneces-

sary information. The algorithms also utilize Pan-

das and Polars for data manipulation and transfor-

mation.

3.3.2 Start and end Trip Schedule

• Start and End Point Extraction Algorithm

This code processes vehicle trip data by identify-

ing start and end points based on key events (key-

on and key-off) before and after vehicle arrivals

and departures at UVA. The process focuses on

merging, ﬁltering, and ranking key-on and key-

off events to capture home locations by calculat-

ing distances between start and end trip locations.

Smart Rideshare Matching: Feasibility of Utilizing Personalized Preferences

485

Algorithm 3: Start-End Trip Points Extraction Algorithm.

1: Input: Arrival, Departure, Key-on, Key-off data, UVA

parking boundary.

2: Output: Excel with start/end trip points, arrival and

departure times, and distances.

3: Load data from Parquet ﬁles into Pandas DataFrames.

4: Convert Timestamp est to timezone-naive datetime.

5: Select Vehicle ID, Timestamp est, Date, Latitude,

Longitude and remove duplicate records.

6: Merge arrival d f with key-on events on Vehicle ID

and Date.

7: Keep the ﬁrst three key-on events before each arrival.

8: Merge departure d f with key-off events on Vehicle ID

and Date.

9: Keep the ﬁrst three key-off events after each departure.

10: Remove trip points within the UVA parking boundary.

11: Use Haversine formula to compute distances between

start to end trip points and arrival to departure points.

12: Merge cleaned arrivals and departures into a f inal d f

with vehicle IDs, dates, trip times, and distances.

13: Export ﬁnal DataFrame as Excel.

3.4 Clustering Approach

The ﬁltered data was further processed to identify po-

tential ride-sharing opportunities. The analysis in-

volved visualizing and clustering vehicles based on

spatial and temporal proximity at various stages of

their trips (Jain and Dubes, 1988). Three distinct clus-

ters were deﬁned to gain insights into the ride-sharing

potential. The clustering logic grouped vehicles by:

• Arrival Points: Vehicles arriving at UVA over the

duration of 10 days composes our arrival points.

This will lead to the potential for detecting ve-

hicles that arrive around the same time. The ar-

rivals spread across UVA boundary over 10 days

can be visualized to show the density of the vehi-

cles around UVA parking lots (Figure 2).

• Departure Points: The location of vehicles leaving

UVA over 10 days captures the departure points.

Again, departure points will help identify the ve-

hicles departing around the same time, offering a

chance to cluster and optimize ride-sharing for re-

turn trips. The departures spread across the UVA

boundary can be visualized as shown in Figure 3.

• Home-to-UVA Points: These points compose our

Home-to-UVA start locations highlighting where

vehicles begin their trips, which is crucial for

identifying potential ride-sharing clusters based

on spatial proximity referring to the vehicles de-

parting for UVA.

• UVA-to-Home Points: Vehicles departing from

UVA to return to home locations will compose

the UVA-to-Home end locations. This will enable

identifying vehicles returning to similar home lo-

Figure 2: Arrivals spread at UVA boundary.

Figure 3: Departures spread at UVA boundary.

cations at similar times, which will be key for op-

timizing ride-sharing on the return trip, reducing

the number of independent return trips.

3.4.1 Deﬁning Clusters

1. Arrival Time Clustering: Vehicles arriving at

UVA within a similar time window (e.g., 15-30

minute intervals) could be grouped together. This

temporal clustering will help identify vehicles that

could potentially share rides to reduce the number

of trips to UVA.

2. Departure Time Clustering: Similarly, vehicles

departing from UVA within the same time win-

dows can be grouped to suggest ride-sharing for

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

486

the return journey. This clustering will help iden-

tify those vehicles that leave UVA at similar times,

enabling ride-sharing for the trip home.

3. Home Location Clustering: This clustering in-

volved identiﬁcation of start and ﬁnal location

of the vehicles. Vehicles that started their trips

from nearby locations (e.g., within 2 miles of each

other) could be identiﬁed as potential ride-sharing

candidates. This was done for both the trip to

UVA and the return trip home. Therefore, vehi-

cles that begin their trips from nearby locations

heading toward UVA or leaving UVA for nearby

locations could be grouped together, suggesting

ride-sharing routes for both legs of the journey.

This analysis allowed us to group vehicles that

start and end their trips from and to nearby locations

respectively within a predeﬁned radius (e.g., 2 miles)

or using a clustering algorithm such as K-means and

arrive at UVA and depart from UVA within the same

time window (Li and Chung, 2020).

4 EVALUATION AND RESULTS

The evaluation of the clustering process was con-

ducted by analyzing the number of vehicles that were

successfully grouped based on the deﬁned criteria.

The results indicate several signiﬁcant clusters of ve-

hicles that could beneﬁt from ride-sharing, especially

during peak arrival and departure times. The analysis

identiﬁed several clusters of vehicles that could po-

tentially beneﬁt from ride-sharing.

4.1 Evaluation Metric 1: Number of

Vehicles in Each Day Within 10

Days Bracket

The provided graphs show the daily number of unique

vehicle arrivals and departures at UVA over a 10-day

period in April 2022. In Figure 4, vehicle arrivals

ranged from a low of 297 on April 15 to a high of 372

on April 22, with some ﬂuctuation observed through-

out the period. In Figure 5, vehicle departures ranged

from a minimum of 285 on April 15 to a maximum of

365 on April 13, indicating a similar level of day-to-

day variability. Overall, both arrivals and departures

exhibit peaks and troughs, but both the distribution

of the unique vehicle arrivals and departures follow a

similar pattern in the researched period.

Figure 4: UVA Unique Arrivals Each Day.

Figure 5: UVA Unique Departures Each Day.

4.2 Evaluation Metric 2: Number of

Vehicles in Each Hour for 10 Days

The following ﬁgures illustrate the hourly distribution

of vehicle arrivals and departures at UVA over the re-

searched 10-day period.

In Figure 6, arrivals have a peak during the morn-

ing hours, particularly between 7 AM and 9 AM,

with the highest number of unique vehicles observed

around 9 AM. After this peak, the number of ar-

rivals gradually decreases throughout the day until

the evening. Figure 7 shows departures, which ex-

hibit a different pattern with peaks occurring in the

late afternoon, particularly between 3 PM and 5 PM,

with a peak at around 4 PM. Departures steadily in-

crease from late morning until reaching their maxi-

mum, after which they decline towards the evening

hours. The data indicates that arrivals are concen-

trated in the morning while departures peak in the late

afternoon, suggesting typical commuting behavior.

4.3 Evaluation Metric 3: Number of

Vehicles in Every 30-Minute

Window for 10 Days

The following ﬁgures contain the number of unique

vehicle IDs arriving and departing at UVA in 30-

minute intervals across the researched 10-day period.

Smart Rideshare Matching: Feasibility of Utilizing Personalized Preferences

487

Figure 6: UVA Unique Arrivals per hour for 10 Days.

Figure 7: UVA Unique Departures per hour for 10 Days.

In Figure 8, the arrivals show a clear peak dur-

ing the morning hours, especially between 7:30 AM

and 9:30 AM, with the highest number reaching 164

around 7:30 AM. After this peak, the number of ar-

rivals steadily declines throughout the day, indicat-

ing that most arrivals happen in the early part of the

day. In Figure 9, the departure reveals a different

trend. Departures increase gradually throughout the

day, peaking between 4:00 PM and 5:30 PM, with the

highest number reaching 182 unique vehicle IDs at

4:30 PM. After this peak, the number of departures

decreases toward the evening hours. This pattern sug-

gests a typical commuting behavior where vehicles ar-

rive in the morning and depart in the late afternoon.

Arrival and Departure Clusters: Signiﬁcant clus-

tering was observed for vehicles arriving and depart-

ing within 30-minute windows, indicating potential

for coordinated ride-sharing as shown in the ﬁgures.

4.4 Evaluation Metric 4: Vehicles

Distribution in Total 10 Days

Table 1 shows the distribution of vehicle counts for

the different number of days starting from 10 days to

9 days and so on, with each number of day’s absolute

count and corresponding percentage of the total. Row

with number of vehicles coming only 1 day out of a

total 10 days has the highest count at 375, making up

38.38% of the total. This may be due to the reason

Figure 8: UVA Unique Arrivals for 10 Days.

Figure 9: UVA Unique Departures for 10 Days.

of having travelers to the hospital or daily visitors to

the University for other purposes. As the total number

of days increases, their respective counts and percent-

ages generally decrease, with total days 10 showing

the lowest count at 76, which accounts for 7.78% of

the total percentage of vehicles. The Grand Total is

conﬁrmed to be comprising 977 total unique vehicle

entries to UVA in 10 days. Similar pattern is seen for

vehicles exiting UVA in these days.

Table 1: Arrival Count Distribution by Vehicle ID and Days.

Total Days Count Percentage

10 76 7.78%

9 46 4.71%

8 35 3.58%

7 36 3.68%

6 46 4.71%

5 46 4.71%

4 72 7.37%

3 92 9.42%

2 153 15.66%

1 375 38.38%

Grand Total 977 100.00%

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

488

5 DISCUSSION

While each day has around 300 vehicles arriving to

and departing from the UVA campus, it is noted that

UVA currently employs about 16,000 faculty and staff

and 7,000 UVA health employees. Given the Wejo

data is only available from newer vehicles made af-

ter 2015 and selected automakers, the study only

matched a small number of vehicles. As this study

focuses on the feasibility of ride-sharing using ve-

hicular telematics data, it is feasible to show ride-

sharing can be matched from these 300 vehicles. It is

possible that 300 vehicles matched daily could have

been over 20,000 vehicles, assuming around 90% of

around 23,000 employees commute by their personal

vehicle. Actual matching would be accomplished via

new sources of data, including ride-sharing App that

UVA plans to adopt in the near future.

6 CONCLUSIONS

This research highlighted the potential to match com-

muters for ride-sharing by leveraging vehicular telem-

atics data. Using the case study of the UVA campus,

we showcased the ride-sharing matching could reduce

trafﬁc and improve transportation efﬁciency for in-

dividuals commuting to and from UVA. Leveraging

both individual preferences in commuting times and

their proximity criteria (i.e., home location and park-

ing lot), this study identiﬁed opportunities for ride-

sharing that could be implemented through a coor-

dinated system, thereby enhancing the overall com-

muting experience while contributing to sustainability

goals. The results suggested that a signiﬁcant num-

ber of trips to and from UVA could be consolidated

through ride-sharing, leading to a more efﬁcient and

sustainable transportation system on campus.

7 FUTURE WORK

Following the analysis of home location clusters, we

will focus on presenting the results of home loca-

tion clusters to explore their implications for optimiz-

ing ride-sharing routes to and from UVA. After ﬁl-

tering and clustering trips that start or end at UVA

and preparing groups of vehicles, route overlap anal-

ysis could provide a deeper probability of providing

ride-sharing opportunities. Evaluating the overlap of

routes for vehicles within the same cluster to suggest

potential ride-sharing pairs or groups.

In addition, future work should also focus on

evaluating a larger, recent dataset to enhance ride-

sharing analysis. Incorporating an expanded dataset,

researchers could capture more travel patterns to ﬁnd

temporal changes in commuting behavior and identify

emerging trends in ride-sharing demand. This would

allow a more comprehensive understanding of current

UVA commuting behaviors and facilitate the design

of more effective ride-sharing initiatives and policies.

ACKNOWLEDGMENTS

This research was funded by the SMARTER Univer-

sity Transportation Center and the Virginia Depart-

ment of Transportation provided data access to the

Wejo data. It is also supported by Science Foundation

Ireland through grant 18/CRT/6222 (ADVANCE).

REFERENCES

de Palma, A., Stokkink, P., and Geroliminis, N. (2022). In-

ﬂuence of dynamic congestion with scheduling pref-

erences on carpooling matching with heterogeneous

users. Trans. Res. Part B: Methodol., 155:479–498.

Hussain, I., Knapen, L., Bellemans, T., Janssens, D., and

Wets, G. (2022). A matching framework for employ-

ees to support carpooling in the context of large com-

panies. IEEE Trans. Intell. Transp. Syst., 23(2):1159–

1170.

Jain, A. K. and Dubes, R. C. (1988). Algorithms for clus-

tering data. Prentice-Hall, Inc., USA.

Li, L., Zhang, H., and Gan, Z. (2023). Factors affecting

college students’ attitudes toward carpooling. Trans-

portation Safety and Environment.

Li, X., Cheng, Y., Peng, H., and Dai, X. (2024). Can

ridesharing improve the reserve capacity of trans-

portation network? IEEE Access, 12:73196–73209.

Li, Y. and Chung, S. H. (2020). Ride-sharing under travel

time uncertainty: Robust optimization and clustering

approaches. Computers & Industrial Engineering,

149:106601.

Masoud, N. and Jayakrishnan, R. (2017). A real-time algo-

rithm to solve the peer-to-peer ride-matching problem

in a ﬂexible ridesharing system. Trans. Res. Part B:

Methodol., 106:218–236.

Narman, H. S., Malik, H., and Yatnalkar, G. (2021). An

enhanced ride sharing model based on human charac-

teristics, machine learning recommender system, and

user threshold time. J. Ambient Intell. Humaniz. Com-

put., 12(1):13–26.

Naumov, S. and Keith, D. (2023). Optimizing the economic

and environmental beneﬁts of ride-hailing and pool-

ing. Prod. Oper. Manag., 32(3):904–929.

OpenStreetMap contributors (2017). Planet dump re-

trieved from https://planet.osm.org . https://www.

openstreetmap.org.

University of Virginia (2023). Updated parkingmastermap.

[Online; accessed October 15, 2024].

Smart Rideshare Matching: Feasibility of Utilizing Personalized Preferences

489