Context-Aware Travel Support During Unplanned

Public Transport Disturbances

Åse Jevinger

, Emil Johansson, Jan A. Persson

and Johan Holmberg

Department of Computer Science and Media Technology, Malmö University, Malmö, Sweden

Keywords: Public Transport, Travel Planner, Context Aware, Prognoses.

Abstract: This paper explores the possibilities and challenges of realizing a context-aware travel planner with

bidirectional information exchange between the actor and the traveller during unplanned traffic disturbances.

A prototype app is implemented and tested to identify potential benefits. The app uses data from open APIs,

and beacons to detect the traveller context (which train or train platform the traveller is currently on).

Alternative travel paths are presented to the user, and each alternative is associated with a certainty factor

reflecting the reliability of the travel time prognoses. The paper also presents an interview study that

investigates PT actors’ views on the potential use for actors and travellers of new information about certainty

factors and travellers’ contexts, during unplanned traffic disturbances. The results show that this type of travel

planner can be realized and that it enables travellers to find ways to reach their destination, in situations where

the public travel planner only suggests infeasible travel paths. The value for the traveller of the certainty

factors are also illustrated. Additionally, the results show that providing actors with information about

traveller context and certainty factors opens up for the possibility of more advanced support for both the PT

actor and the traveller.

1 INTRODUCTION

Smartphones create new opportunities within Public

Transport (PT) to both provide the traveller with more

personalized information, and give the traveller

possibilities to share their own information with the

PT actor. For instance, information shared by the

traveller about where the traveller currently is located

can be used by the PT actor to improve transport

services (Stelzer et al., 2016). If this type of

bidirectional information exchange is used for

transmitting real-time information during unplanned

traffic disturbances, both the passengers’ decision

basis for the continued journey and the PT actors'

decision basis for disturbance management can be

improved. Previous research shows that information

about disturbances and alternative travel paths is

highly prioritized by the travellers (Currie och Muir,

2017; Hörold et al. 2014). Moreover, the travellers

prefer to get this information on an individual level

(i.e. specified according to the individual travel plan),

rather than on aggregated level (for instance, an

https://orcid.org/0000-0002-6019-1182

https://orcid.org/0000-0002-9471-8405

overview of all traffic disturbances within an area)

(Hörold et al. 2014). Previous research also shows

that information from the traveller has the potential to

improve the PT actors’ decisions and actions,

especially in traffic management (Mayas et al. 2015).

The primary aim of this study is to investigate

potential benefits for actors providing PT services and

travellers, of travel/journey planners that are context-

aware och enable bidirectional information exchange

between the actor and the traveller, during unplanned

traffic disturbances. A secondary aim is to explore the

possibilities and challenges of realizing such a travel

planner, including demonstrating potential benefits,

through a prototype implementation. The main

purpose with the study is to increase knowledge of

how the support for both travellers and PT actors can

be improved, when decisions have to be made due to

unplanned traffic disturbances. Thereby, a more

attractive public transport and increased system

efficiency may be obtained (e.g. through more

efficient disturbance management and improved

resource utilization).

160

Jevinger, Å., Johansson, E., Persson, J. and Holmberg, J.

Context-Aware Travel Support During Unplanned Public Transport Disturbances.

DOI: 10.5220/0011761000003479

In Proceedings of the 9th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2023), pages 160-170

ISBN: 978-989-758-652-1; ISSN: 2184-495X

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

The study includes three parts: interviews with PT

actors, development of a prototype in the form of an

app, and app tests in different scenarios. The

interviews investigate the potential use for actors and

travellers of new information about certainty factors

associated with travel times, as well as the potential

use for actors of information about travellers’

contexts and travellers’ destinations, based on

bidirectional communication with the traveller (e.g.

via an app). The app development part investigates if

and how this type of context-aware travel planner can

be developed, and enables scenario tests. It also

explores how certainty factors associated with travel

times can be developed and implemented in a travel

planner. Finally, the scenario tests are used to study

the effects in travel time and travel alternatives

presented to the traveller, when using the existing app

versus the new prototype app during unplanned

disturbances in the south of Sweden. The interviews

encompass different types of traveller context

information; thereby, they cover a broader area than

the app development and tests, as the app only

considers contextual information related to which

train or train platform the traveller is currently on, in

combination with traveller destination.

The prototype app is developed independently

from current travel planner apps, but its functionality

is, in a longer perspective, intended to be integrated

with existing travel planners. The app operates as

follows. In the event of a traffic disturbance, the

traveller is given the opportunity to request

personalized information through the app. This

request also means that the traveller agrees to share

information with the PT actors about his/her desired

destination and current context (i.e. which train or

train platform the traveller is currently on). The

destination is provided manually by the traveller

whereas the context is detected by the app, i.e. it

detects current train or platform instead of the

geographical position, which is common for regular

travel planners. This means that the alternative travel

paths provided to the traveller can be based on future

train stops, instead of the train stations or bus stops

that the train is currently close to but that may be

impossible to reach due to the itinerary of the train.

Based on the traveller's destination and context, as

well as disturbance information and timetables from

open APIs, the app provides suggestions for

alternative ways to continue the journey. Each travel

alternative is associated with a certainty factor

reflecting the reliability of the travel time prognosis,

based on time and the reason for the disturbance. The

travellers can thus choose between the different travel

alternatives based on the certainty of the different

travel alternatives and their personal needs. Thereby,

a more informed decision about continued travel can

be made. This type of certainty factor can also serve

as a support for traffic managers and traffic

informants.

The main differences between the focus of this

study and the existing solutions commonly used today

are:

1. The prototype app is context aware and is

thereby able to provide the traveller with personalized

information about alternative travel paths, in

particular during unplanned traffic disturbances (e.g.

advising the traveller to alight at the next stop and

then catch another bus/train).

2. PT actors get access to information about

the travellers’ contexts and their destinations, which

can be used in disturbance management.

3. Travellers and PT actors get access to

certainty factors, which are based on time and the

reason for the disturbance, for the travel time

prognoses. These certainty factors can be used both in

disturbance management and by the travellers to

make more informed decisions during unplanned

traffic disturbances.

The remainder of the paper is organized as

follows. Section 2 provides an overview of the related

work. Section 3 describes the methodology used for

the interviews, the prototype development and the

scenario tests. Section 4 presents the results and in

section 5, the conclusions are drawn and discussed.

2 RELATED WORK

Over the years, travel planners have evolved from

only presenting static information to the traveller, to

also including real-time and predicted information.

This development can be mainly attributed to the

widespread use of smartphones and technological

advancements. In particular, the introduction of real-

time information related to traffic disturbances has

been of great importance to travellers. However,

there are also other types of real-time information that

can be incorporated into a travel planner. For

instance, Georgakis et al. (2020) present the

implementation of a travel planner for MaaS schemes

that includes dynamic constraints (e.g., available on-

demand service offers with different estimated time

of arrivals).

There are a few travel planners on the market that

try to estimate the traveller context. For instance,

Czech train operator České Dráhy has developed a

“context-aware” app that suggests alternative routes

during disturbances, in order to ease train travel

Context-Aware Travel Support During Unplanned Public Transport Disturbances

161

during the upgrade of the Czech railway system to

ETCS. The app relies on purchased tickets to discover

with which train the passenger travels (Tomášek

2021). Within research, there are several research

studies presenting different methods for detecting a

traveller’s transport mode (Sadeghian et al. 2021;

Stenneth et al. 2011). There are also studies that

predict the travellers’ contexts on an aggregate level.

For instance, Benchimol et al. (2021) predict future

passenger flows at stations and onboard trains. This

information is then used by a travel planner that

incorporates the predictions as criteria, meaning that

the predictive load information is not only shown to

the traveller as information, but also used as a search

criterion when trip alternatives are identified by the

travel planner. Most studies focusing on passenger

flow predictions use data from different types of local

readers, cameras or sensors (e.g., automated fare

collection or automatic passenger counting

technologies). However, other types of data have also

been investigated, for instance, from social media

(Zhu et al. 2017).

As for information exchange between the

operator and the traveller, connected to travel

planning support, it is common to only use the GPS

position and personal preferences. For instance,

Esztergár-Kiss (2019) provides an overview of

European travel planners, with a particular focus on

multimodality, and presents a framework for

evaluation of such travel planners. Notably, none of

the studied travel planners include travel context

beyond the personal preferences related to the trip

(e.g., maximal number of transfers or preferred

transport mode) and GPS coordinates. However,

some conceptual work has been done to identify what

context information related to a traveller can be used

for improving travel support. In particular, Jevinger

and Persson (2019) investigate how the traveller’s

current context can be utilized in the travel planner to

provide better support during unplanned

disturbances.

Thus, we have failed to find previous research that

utilizes context information in terms of which train or

train platform the traveller currently is on, to improve

the travel planner.

There are a lot of research studies on different

ways of predicting arrival times and replanning traffic

in case of delays within PT (e.g., Josyula (2020), Xu

and Ying (2017)), but significantly fewer studies on

how to estimate (and communicate) the certainty of

such predictions. Many of the studies that do exist,

base their estimations on statistical methods

(O'Sullivan et al. 2016; Rahman et al. 2018). In

particular, O'Sullivan et al. (2016) have an approach

that is similar to ours. They view estimated arrival

times from traveller information systems as black

boxes, and calculate prediction intervals based on

historical data of estimated arrival times, combined

with actual outcome (i.e., the actual arrival times).

However, their study does not take the causes of the

traffic disturbances into account.

Studies with other objectives than estimating

prognosis certainty, often make statistical

assumptions on the certainty to illustrate something

else (e.g., how such certainties can be communicated

or used in travel planning) (Botea and Braghin, 2015;

Fernandes et al. 2018). For instance, Botea and

Braghin (2015) assume that the certainty of arrival is

normally distributed with mean value 0 and standard

deviation 80 seconds or 40 seconds. This is then used

to develop stochastic itineraries that, unlike ordinary

itineraries with sequentially arranged transport

routes, may include several route alternatives, taking

into account potentially missed connections.

Naturally, there are also several studies that apply

optimization or machine learning to estimate both

arrival time and certainty (Coffey et al., 2011; Yu et

al., 2017). For instance, Yu et al. (2017) use

Relevance vector machine to estimate the arrival time

of buses, as well as lower and upper bounds for this,

based on confidence intervals. The outcome is then

compared with five other traditional machine learning

methods.

In summary, to the best of our knowledge there is

a lack of understanding of how a travel planner that

uses information about the traveller’s context can be

developed and what the potential benefits may be.

There is also a need for increased understanding of

the potential use for actors and travellers of new

information related to certainty factors, travellers’

contexts and travellers’ destinations. Additionally,

while there are studies estimating certainty in arrival

time prognoses in PT, we have failed to find studies

that base these estimations on the actual disturbance

reasons. This study aims to address these research

gaps.

3 METHODOLOGY

3.1 Interviews with PT Actors

The interviews focused on what information is

available today about the traveller's context and

destination, and how certainty in travel time

prognoses is calculated, as well as how new

information about these things could be used and

affect travellers and actors (see interview questions in

VEHITS 2023 - 9th International Conference on Vehicle Technology and Intelligent Transport Systems

162

Appendix). The interviews were semi-structured, and

since they took place during the Corona pandemic,

they were conducted over a digital communication

platform (Microsoft Teams).

The following officials were interviewed:

 Organizational developer within rail traffic

information, at the Swedish Transport

Administration

 Project manager within traffic information, at

the Swedish Transport Administration

 IT architect with focus on traffic information,

at Skånetrafiken (the regional public transit

authority in the south of Sweden)

 Organizational developer within traffic

information, at Skånetrafiken

 Head of travelling staff, at SJ Öresund (a

Swedish railway company operating in the

south of Sweden)

 Traffic informers (two persons), at the Swedish

Transport Administration

The IT architect and the Organizational developer

at Skånetrafiken were interviewed together, and the

Traffic informers were interviewed separately, i.e. in

total 6 interviews were conducted. Two researchers

asked questions and took notes during each interview.

After each interview, the notes were compiled and

sent to the respondent for validation. The majority of

the respondents returned with a confirmation of

correctness, while a few returned with clarifications

and corrections, which were added to the

compilations.

3.2 Prototype Development

There are different ways to detect which train or

platform a traveller is currently on, that do not require

any active participation from the traveller. Today’s

travel planners often detect the traveller’s physical

position through GPS, and then search for nearby PT

stops. This solution could be used to detect the train

station a traveller is currently on; however, for the

onboard and platform situations, other solutions are

necessary. One way for the onboard situation is to

map the train GPS position with the traveller’s GPS

position. By using an algorithm for this type of GPS

mapping from previous research (see e.g., Stenneth et

al. 2011), the current train could be revealed. In the

south of Sweden, many trains are equipped with GPS;

however, depending on operator, the open APIs lack

real-time positioning data for some of them.

Moreover, there might be a time delay between the

train positioning data and the traveller positioning

data, which may cause mapping problems.

Another solution may be to use a personal area

network technology to detect which train a traveller

is currently on, for instance, Bluetooth from fixed

beacons installed onboard the train or a wifi network

open to the passengers onboard a train. With

appropriate software, the MAC address of an onboard

wifi router could be identified, and with a mapping

between trains and MAC addresses, the current train

could be detected. However, such a mapping is not

available today in the Swedish open APIs.

Alternatively, if the wifi transmitted information

about the current train, the app could easily just pick

this information up. However, this type of

information is not available as a pure data string.

Another problem that concerns both GPS and wifi is

that it might be difficult to separate two trains

standing next to each other.

In this study, we chose to use fixed Bluetooth

Low Energy (BLE) beacons. The beacons

broadcasted the physical train ID or platform ID,

which was picked up by the prototype app. The

physical train ID was mapped to the line the train was

currently running on (using the open APIs). The

advantages with this solution are that, thanks to the

short coverage of BLE, it is easier to separate two

nearby trains. Moreover, the open APIs do contain a

mapping for many operators, which makes this

solution possible. The drawbacks are that beacon

installations and battery changes/power cord

installations are required, which entails costs.

For the calculation of certainty factors, the

Swedish Transport Administration provided us with

three sets of traffic disturbance data covering three

months: July 2021, December 2021 and February

2022. Thereby, potential season variations were

accounted for. The data included reason code (i.e., the

cause of the disturbance), announced time (i.e., time

of departure/arrival according to timetable), reporting

time (i.e., time when new prognosis was announced)

and the error margin of the prognosis (i.e., difference

between prognosis and actual departure/arrival as an

absolute value). As our access to disturbance data was

limited, while the data needed to be extensive enough

to provide credible results, the study only focused on

the reason codes that appeared more than 1000 times

in the data files. For the others, only the mean value

of the error margin of the prognoses were calculated.

Initial regression analyses on the different variables

above showed significant relations between reason

code, error margin of the prognosis, and time interval

between reporting time and announced time.

Therefore, we chose to focus the certainty factor

calculations on these variables. The calculations

included mean values of the error margin of the

Context-Aware Travel Support During Unplanned Public Transport Disturbances

163

prognoses, given different time intervals between

reporting time and announced time, together with

prediction intervals and determination coefficients. In

those cases where the lower bound of a prediction

interval was negative, it was set to zero, since the

error margin is at least zero (we only focus on the

difference between prognosis and actual

departure/arrival as absolute values).

3.3 Scenario Tests

The scenario tests focused on disturbances in the rail

traffic between Lund central station and Malmö

central station, in the south of Sweden. The main

reason for selecting this stretch of track is that it is

considered a bottleneck in Swedish rail traffic. A

large number of travellers (and freight transports) use

this stretch daily (up to 60,000 travellers per day

according to Skånetrafiken (2020)), and unplanned

disruptions thereby create major problems. In

addition, there is a relatively large range of alternative

travel routes around the stretch.

During a period of four weeks, four researchers

manually monitored the traffic between these two

stations. For all major disturbances, searches for trips

were made using both the existing public travel

planner and the new prototype app.

4 RESULTS AND ANALYSIS

4.1 Interviews with PT Actors

The findings from the interviews related to the

information available today, can be summarized as

follows.

Regarding the currently available information

about the traveller's destination and context, the

results show that the actors work with estimates of the

number of travellers on trains, and that they have

relatively little information about the traveller's

destination and other contexts. The traveller estimates

are often based on the onboard staff's manual

calculations, onboard ticket validations or registration

of activated tickets in the travel planner app, and they

are used to provide the traveller with congestion

forecasts or to facilitate evacuation, if needed. The

information from travel tickets often reveals which

zones a traveller is crossing, but usually not which

route and departures the traveller has chosen. The

Swedish Transport Administration also has cameras

installed on certain train platforms, which are

activated when needed. These cameras are used to,

for instance, verify that the travellers seem to have

understood, and thus reacted to, certain traffic

announcements, e.g., about track change. The

cameras are sometimes also used to manually

estimate how many travellers are on a platform.

Information about travellers in wheelchairs or with

bicycles/wheelchair/etc. may be manually collected

through observation by the onboard staff (unless a

special seat ticket, e.g., for a wheelchair, has been

booked). This information is sometimes transmitted

via telephone to the train dispatcher, for traffic

planning purposes.

Regarding how certainty in travel time prognoses

is calculated today, the results show that the Swedish

Transport Administration has relatively good

knowledge of which prognoses are certain and which

are uncertain. However, the certainties are not

quantified, but primarily based on the personal

experiences of the staff. For instance, it is often

possible to state with greater reliability when a

missing train driver will arrive, than to state how long

will it take for traffic to recover from people illegally

crossing rail tracks (which usually involves police

action). As a consequence of the uncertainty, the

actors are a bit restrictive in communicating less

reliable prognoses to the traveller. There is a risk that

the traveller perceives prognoses as promises, and

when they are broken, frustration and demands for

compensation may arise. In order to give some

expression to certainty in prognoses, a number of

keywords are used in the communication with the

traveller, e.g., "departs at the earliest", “preliminary

time” and “await time” (used when awaiting a time

when the train will run again). The keyword will be

removed when there is a more reliable time prognosis.

The findings from the interviews related to how

new information about certainty in travel time

prognoses, traveller's context and traveller's

destination, could be used and affect travellers and

actors, are shown in Table 1.

In summary, the interviews show that the PT

actors today work with estimates of the number of

passengers on trains, and their knowledge of which

prognoses are certain or uncertain are primarily based

on the personal experiences of the staff. Information

about the travellers’ destinations can only be obtained

for those who have chosen to buy a ticket that

specifies the entire journey. This means that

information about the commuters’ destinations is

usually not available. In addition, the actors lack other

contextual information beyond what can be visually

observed by the onboard staff, e.g., different types of

disabilities. The information at hand for the on-board

staff is only to a limited extent distributed to relevant

actors.

VEHITS 2023 - 9th International Conference on Vehicle Technology and Intelligent Transport Systems

164

Table 1: Interview results.

Information Use and effects

Traveller’s

destination

and the train

or train

platform the

traveller is

currently on

Traffic track planning (to minimize

distance and time for travellers when

chan

trains

)

Train prioritizations (e.g., prioritizing a

full train over a relatively empty train)

Estimation of time for boarding and

alighting, which can be used for early

rediction and communication of dela

Improved train-replacement traffic

planning in terms of route and vehicle

capacity, e.g. taxi to places that few

travellers go to and many direct large

uses to

laces that man

travelle

o to

Improved evacuation

Long-term traffic planning

Quantified

certainty in

travel time

prognoses

Improved support to the traveller in

assessin

different trans

ort alternatives

Earlier prognoses provided to the

travellers (as they will not interpret them

as promises)

Improved actor support when

communicating disturbances to travellers

(as the certainty is quantified)

Facilitated work for new employees with

little experiences of estimating certainties

Improved planning of staff (e.g. customer

service) and activated communication

channels (during large disturbances)

Many

travellers with

many/big

suitcases

Traffic track planning (to minimize

distance and time for travellers when

changing trains)

Estimation of time for boarding and

alighting, which can be used for early

rediction and communication of dela

Travellers

with bicycles/

wheelchair/

etc.

Early information to the traveller about

lack of space on planned train (requires

technology for detecting empty onboard

space/seats)

Ordering of train-replacement traffic that

allows for bicycles/

wheelchair/etc.

Traffic track planning (to minimize

distance and time for travellers when

changing trains, and to have a working

elevator or no stairs)

Estimation of time for boarding and

alighting, which can be used for early

rediction and communication of dela

Traveller

disabilities

(e.g. vision or

hearing

impairment)

Improved support both during

disturbances and in normal situations (e.g.

help with finding correct seat, information

about which trains or carriages are less

full, verification that traveller has boarded

the correct train)

roved evacuation

With the help of the app described above,

information about a traveller’s destinations and

whether the traveller is on a particular train or

platform, can be obtained. Table 1 shows that this

type of information, together with certainty factors

for alternative travel routes, can be useful for a range

of different situations, for both travellers and actors.

4.2 Prototype

4.2.1 Architecture and Design

Based on the detected train or train platform that the

user is currently on, as well as the end destination as

specified by the user, several travel plans are

generated. These are found using the Resrobot Route

Planner API, an open API that includes all Swedish

PT operators. If the user is on a train, travel plans are

created from every upcoming station for that train. If

the user is standing on a platform, travel plans are

simply created with that station as the origin. In order

to also find suppressed travel paths, i.e., paths that

according to the original timetables have longer travel

times than the most time-efficient travel paths,

different manipulations, such as excluding trains in

the search, were applied. If the used traveller planner

API would have used the most updated prognoses

when finding the travel paths, this would not be

necessary. The found travel plans are then updated

with data on delays and cancellations in the train

traffic, received from the open API provided by the

Swedish Transport Administration (“Trafikverket”).

If any leg in a travel plan is cancelled, or its arrival

time is delayed beyond the departure time of the next

leg (assuming a three-minute minimum transit time

between legs), that travel plan is filtered away. Due

to a lack of real-time data on delays and disturbances

for other modes of transports than train, only

timetable times are used for transport legs with bus.

Of the remaining plans, the ones with an arrival time

at the final destination that is closest in time, are

presented to the user. The app allows the user to save

one of these plans, so that it can be viewed later even

when the user is not on a train or platform.

The server side of the application was

implemented in Java as a Spring Boot application,

while the front-end was created using the Apache

Cordova framework and implemented in Javascript,

HTML and CSS. Figure 1 shows the final system

architecture and Figure 2 shows how the app operates.

Context-Aware Travel Support During Unplanned Public Transport Disturbances

165

Figure 1: System architecture of the app prototype.

Figure 2: Flow chart illustrating how the app operates.

4.2.2 Certainty Factors

As mentioned above, this study only focused on the

reason codes that appeared more than 1000 times in

the provided data files. This criterion was fulfilled by

11 disturbance reason codes (see Table 2). These

reason codes represent disturbances related to

accidents, unexpected dwelling times, delayed

connecting trains, priorities of other trains, track

errors and other infrastructure failures. The regression

analysis on the error margin of the prognosis in

relation to the time interval between reporting time

and announced time, for the different reason codes,

showed that the F values and the P values were less

than 10

-11

for all the reason codes. This means that

there is a clear correlation between the error margin

of the prognosis and the time interval between

reporting time and announced time. Table 2 shows the

results of the statistical analysis. Column three shows

the determination coefficients (r

), columns 4 to 7

show the mean error margins of the prognoses and the

corresponding 80% prediction intervals, when the

time intervals between the reporting time and

announced time are 10 min. and 40 min., respectively.

The determination coefficient indicates how much of

the variation in the error margin can be explained by

the different time interval between reporting time and

announced time, whereas the 80% prediction

intervals are intervals within which the error margins

lie with an 80% probability, given the different time

intervals between the reporting time and the

announced time.

As can be seen, the prediction intervals are

different for different reason codes. For ONA – and

OUT 01, the intervals are 23.88 and 10.32 minutes,

respectively, when the time intervals between the

reporting time and announced time are 10 min. The

same pattern can be seen when the time intervals

between the reporting time and announced time are

40 min. This means that the reason code, i.e., the

reason for the disturbance, has a relatively large

impact on how certain a given prognosis is. Access to

information about the reason code can thus provide

support in estimating the reliability of prognoses

during unplanned disturbances.

Table 2: Statistical analysis of certainty factors.

Reason

code

#data

entries

Deter.

coeff.

10 min. report-

ann.

40 min. report-

ann.

Mean Pred. Mean Pred.

DPR 03 8091 0.13 3.83 0-10.72 5.82 0-12.70

DPS 01 1610 0.08 4.30 0-12.68 6.05 0-14.43

IBÖ 01 2263 0.25 5.46 0-17.83 7.39 0-19.77

IBÖ 02 1368 0.10 4.54 0-10.89 6.57 0,21-

12.92

JTP - 1817 0.04 4.37 0-16.67 6.59 0-18.89

JTP 13 1066 0.13 3.74 0-17.14 7.11 0-20.52

OMÄ 02 1885 0.11 6.14 0-13.97 8.56 0.73-

16.40

OMÄ 03 1087 0.10 4.99 0-11.52 6.69 0.15-

13.23

ONA - 1452 0.11 9.61 0-23.88 13.35 0-27.62

OUT - 3043 0.06 5.09 0-15.15 7.20 0-17.27

OUT 01 2446 0.18 4.68 0-10.32 8.63 2.99-

14.28

A comparison between columns 5 and 7 in Table

2 shows that the prediction intervals are slightly

larger when the time intervals between the reporting

time and announced time are 40 min. than when they

are 10 minutes. This result indicates, together with the

determination coefficients in Table 2, that some part

of the variations in the error margin can be explained

by the time interval between the reporting time and

the announced time. This means that access to

information about the time interval between the

reporting time and the announced time can also

provide support in the estimation of the reliability of

prognoses during unplanned disruptions.

4.3 Scenario Tests

Figures 3 and 4 show an example of the differences

between the public travel planner offered today to

people traveling in the south of Sweden, and our

prototype app, when an unplanned traffic disturbance

means that the traveller is likely to miss the next

Timetables,

Prognoses

Context,

Destination

Alternative routes,

travel times with

estimated certainties

Traveller contexts and

destinations,

Travel times with estimated

certainties

VEHITS 2023 - 9th International Conference on Vehicle Technology and Intelligent Transport Systems

166

connection. Figure 3 shows the interface of the public

travel planner when searching for a trip between two

cities (Kritstianstad C and Svedala station) during this

traffic disturbance. In this scenario, the traveller has

already boarded the train 1103 from Kristianstad C.

The app suggests that the traveller changes train in

Lund C. However, train 1103 is expected to arrive to

Lund C after the train to Svedala has already departed

(at 20:29 and 20:25 respectively). Thereby, this

itinerary will probably not be feasible.

Figure 3: Public travel planner interface when searching for

a trip from Kristianstad C to Svedala station, during an

unplanned disturbance.

Figure 4: The suggested alternative travel paths in the

prototype app, during the unplanned disturbance shown in

Figure 3.

Figure 4 shows the interface of our prototype app

during the same traffic disturbance. This app presents

three alternative itineraries with different certainty

factors. Alternative 1 suggests switching from train to

bus 165 in Lund C. This bus departs 7 minutes after

the train's estimated arrival in Lund. Based on

historical data on certainty in prognoses for this type

of disturbance, the calculated certainty factor for this

travel alternative is 0.715. This means that at the time

of the search, the probability that the sum of the given

prognosis and the prediction interval is less than the

time available for changing vehicles (including 3

minutes for walking between vehicles) is 0.715. If it

had been closer to 1, choosing this alternative would

have been safer. Alternative 2 suggests changing to

train 1287 in Lund C. Alternative 3 also suggests

switching to train 1287, but in this case in Malmö C

instead. The certainty factor becomes higher in both

of these alternatives, since they involve longer

exchange times between arrival and subsequent

departure.

Figure 5: Public travel planner interface when searching for

a trip from Östra Grevie station to Lomma station, during

an unplanned disturbance.

Figures 5 and 6 show an example of the

differences between the public travel planner and our

prototype app, when an unplanned traffic disturbance

causes upcoming stops of the train the traveller is on,

to be cancelled. In this scenario, the traveller is

currently on train 1716 and is going from Östra

Grevie station to Lomma station. Figure 5 shows that

the public travel planner suggests traveling via

Malmö C. However, since the train is cancelled for

the last stops, it will never arrive at Malmö C. This

itinerary will thereby not be feasible. Our prototype

app shows three alternative travel routes, see Figure

6. All travel routes include train 1420 and their

certainty factors are of the same magnitude. What

Context-Aware Travel Support During Unplanned Public Transport Disturbances

167

separates the travel routes is where the traveller gets

off the current train and gets on train 1420.

Figure 6: The suggested alternative travel paths in the

prototype app, during the unplanned disturbance shown in

Figure 5.

The above examples show how the prototype app,

through knowledge of the traveller's destination and

context (in this case which train the traveller is

currently on), can provide an updated itinerary that is

difficult for the traveller to identify using the public

travel planners. The examples also illustrate the value

for the traveller of providing certainty factors

connected to different travel alternatives.

5 DISCUSSION AND

CONCLUSIONS

This paper has shown that a travel planner detecting

the traveller’s context has the potential to provide

better support for both the traveller and PT actor

(given that the traveller is willing to share context

information), by extracting relevant information that

enable better decision making, during unplanned

disturbances. Furthermore, the potential of using

disturbance reason codes to make better estimates of

the reliability of the travel time prognoses, has also

been shown.

The interviews with PT actors showed that the

actors today work with estimates of the number of

passengers on trains, and that their knowledge of

which prognoses are certain or uncertain are primarily

based on the personal experiences of the staff.

Information about the travellers’ destinations can

only be obtained for those who have chosen to buy a

ticket that specifies the entire journey, i.e., this

information is usually not available for commuters.

The interviews also showed that providing the actors

with information about the travellers’ destinations,

which train or train platform a traveller is currently

on, and other types of contextual information (such as

disabilities), opens up for the possibility of more

advanced support for both the PT actors and the

travellers. Certainty factors associated with different

alternative travel paths may also be of value for both

actors and travellers.

The prototype development showed how a travel

planner with bidirectional information exchange

between the actor and the traveller, that is aware of

the traveller’s context (in this study which train or

train platform the traveller is currently on), and that

presents alternative travel paths with associated

certainty factors during traffic disturbances, can be

realized. In particular, the study showed how these

certainty factors can be calculated based on

information about the reason for the traffic

disturbance.

Finally, the paper illustrated what types of

benefits such a travel planner may provide for the

traveller. The updated itineraries presented for the

user takes into account the traveller’s personal

context and destination, and suggests travel paths that

are not provided by the public travel planner used

today. In particular, the study showed that with this

new type of travel planner, the traveller can find ways

to reach the destination, when the public travel

planner only suggests infeasible travel paths. The

value for the traveller of different certainty factors

connected to the travel alternatives was also

illustrated.

As mentioned in Section 4.2, there is a lack of

open real-time data on delays and disturbances for

other modes of transports than train, in the south of

Sweden. Thereby, the certainty factors and the

context awareness developed in the travel planner

prototype, only focus on trains. Ideally, it should

incorporate more transport modes. However, the

methods developed in this study for estimating

certainty factors and for obtaining context awareness

may be applied to other PT transport modes as well,

if data can be made available.

The new type of travel planner presented in this

study may be used by the PT actor to collect

information about the amount of passengers onboard

different trains or on different train platforms.

However, the collected data will only represent a

VEHITS 2023 - 9th International Conference on Vehicle Technology and Intelligent Transport Systems

168

fraction of the passengers, since all may not be

interested in using the travel planner. This fraction

can potentially be estimated based on how many

travellers use the travel planner in general, which may

enable a more stable number of passengers onboard

or at platforms. However, the more travellers who

choose to use the app, the more reliable information

can be achieved.

Furthermore, this study has not considered any

additional costs or tickets that might be needed when

switching routes within the public transport network.

In the south of Sweden, zone-based tickets are often

applied, which means that this is usually not a

problem. However, in other context this may need to

be considered.

For future studies, we believe it would be

interesting to investigate how other transport modes

can be added (e.g. taxi, bicycle, bus). It would also be

interesting to study how the context awareness could

be expanded to other environments (e.g. home, on the

way to a bus stop).

ACKNOWLEDGEMENTS

This study has been funded by the Swedish Transport

Administration.

REFERENCES

Benchimol, P., Amrani, A., Khouadjia, M. (2021,

September). A Multi-Criteria Multi-Modal Predictive

Trip Planner: Application on Paris Metropolitan

Network. In 2021 IEEE International Smart Cities

Conference (ISC2), 1-4, IEEE.

Botea, A., Braghin, S. (2015). Contingent versus

deterministic plans in multi-modal journey planning. In

International Conference on Automated Planning and

Scheduling, 25, 268-272.

Coffey, C., Pozdnoukhov, A. Calabrese, F. (2011). Time of

arrival predictability horizons for public bus routes. In

ACM SIGSPATIAL International Workshop on

Computational Transportation Science, 1-5.

Currie, G., Muir, C. (2017). Understanding passenger

perceptions and behaviors during unplanned rail

disruptions. Transportation research procedia, 25,

4392-4402.

Esztergár-Kiss, D. (2019) Framework of Aspects for the

Evaluation of Multimodal Journey Planners.

Sustainability, 11, 4960.

Fernandes, M., Walls, L., Munson, S., Hullman, J., Kay, M.

(2018). Uncertainty displays using quantile dotplots or

cdfs improve transit decision-making. In CHI

Conference on Human Factors in Computing Systems,

1-12.

Georgakis, P., Almohammad, A., Bothos, E., Magoutas, B.,

Arnaoutaki, K., Mentzas, G. (2020). Heuristic-based

journey planner for mobility as a service (MaaS).

Sustainability, 12(23), 10140.

Hörold, S., Mayas, C., Krömker, H. (2014). Passenger

needs on mobile information systems–field evaluation

in public transport. In Advances in Human Aspects of

Transportation: Part III, AHFE Conference, pp. 115-

124.

Jevinger, Å., Persson, J.A. (2019). Potentials of Context-

Aware Travel Support during Unplanned Public

Transport Disturbances. Sustainability, 11(6), 1649.

Josyula, S.P., Krasemann, J.T., Lundberg, L. (2020).

Parallel computing for multi-objective train

rescheduling. IEEE Transactions on Emerging Topics

in Computing, 9(4), 1683-1696.

Mayas, C., Hörold, S., Stelzer, A., Englert, F., Krömker, H.

(2015). Evaluation of dispatcher requirements on

automated customer feedback in public transport. In

IFIP Conference on Human-Computer Interaction,

537-541, Springer.

O'Sullivan, A., Pereira, F.C., Zhao, J., Koutsopoulos, H.N.

(2016). Uncertainty in bus arrival time predictions:

Treating heteroscedasticity with a metamodel

approach. IEEE Transactions on Intelligent

Transportation Systems, 17(11), 3286-3296.

Rahman, M.M., Wirasinghe, S.C., Kattan, L. (2018).

Analysis of bus travel time distributions for varying

horizons and real-time applications. Transportation

Research Part C: Emerging Technologies, 86, 453-466.

Sadeghian, P., Håkansson, J., Zhao, X. (2021). Review and

evaluation of methods in transport mode detection

based on GPS tracking data. Journal of Traffic and

Transportation Engineering (English Edition), 8(4),

467-482.

Skånetrafiken (2020). Lyckat tågstopp mellan Malmö och

Lund. https://www.mynewsdesk.com/se/skanetrafiken/

pressreleases/lyckat-taagstopp-mellan-malmoe-och-lu

nd-3030112 [2022-11-15]

Stelzer, A., Englert, F., Hörold, S., Mayas, C. (2016).

Improving service quality in public transportation

systems using automated customer

feedback. Transportation Research Part E: Logistics

and Transportation Review, 89, 259-271.

Stenneth, L., Wolfson, O., Yu, P. S., Xu, B. (2011).

Transportation mode detection using mobile phones

and GIS information. In Proceedings of the 19th ACM

SIGSPATIAL international conference on advances in

geographic information systems, 54-63.

Tomášek, V. (2021). Jízdenku na zpožděný spoj zatím

aplikace Můj vlak nenabízí. České dráhy to ale chtějí

změnit. iROZHLAS, August 29. https://www.irozhl

as.cz/veda-technologie/technologie/muj-vlak-aplikace-

ceske-drahy-zpozdene-spoje-jizdenka_2108291605_ako

[2022-11-15]

Xu, H., Ying, J. (2017). Bus arrival time prediction with

real-time and historic data. Cluster Computing, 20(4),

3099-3106.

Yu, H., Wu, Z., Chen, D., Ma, X. (2016). Probabilistic

prediction of bus headway using relevance vector

Context-Aware Travel Support During Unplanned Public Transport Disturbances

169

machine regression. IEEE Transactions on Intelligent

Transportation Systems, 18(7), 1772-1781.

Zhu, S., Wang, Y., Shang, S., Zhao, G., Wang, J. (2017).

Probabilistic routing using multimodal data.

Neurocomputing, 253, 49-55.

APPENDIX

The interviews were conducted as follows. First, the

project was introduced for the respondents, and the

concept of uncertainty factors was explained. Then,

the respondents were asked to describe their previous

and current professional roles and associated tasks.

Thereafter the following questions were asked:

1. Concerning the information about the travellers

that is available for you today:

a. What information do you have about the

traveller’s destination and which train or

train platform a traveller is currently on?

b. What other types of traveller context

information do you have access to?

2. If you do not have access to information about

the traveller’s destination and which train or train

platform a traveller is currently on:

a. How could such information be used?

b. What effects do you think this would have

on travellers and PT actors?

3. Concerning any additional information about the

traveller’s context that you would like to have

access to:

a. How could such information be used?

b. What effects do you think this would have

on travellers and PT actors?

4. Concerning information corresponding to

uncertainty factors in prognoses that is available

for you today:

a. What type of information do you have

access to, if any?

b. How is this information used?

5. If you do not have access to information

corresponding to uncertainty factors:

a. How could such information be used?

b. What effects do you think this would have

on travellers and PT actors?

VEHITS 2023 - 9th International Conference on Vehicle Technology and Intelligent Transport Systems

170