Page 62 - Proceeding of Atrans Young Researcher's Forum 2019_Neat

“Transportation for A Better Life:
Smart Mobility for Now and Then”

23 August 2019, Bangkok, Thailand

One of the pioneers of developing apps for The absence of travel mode information in
collecting GPS data was the team of F. Zhao et al. GPS data has prompted researchers to develop
[4] who conducted Future Mobility Sensing (FMS) solutions to infer how participants traveled. There
during 2012-2013 as a sub-project of the Household are three main groups, including deterministic
Interview Travel Survey (HITS) in Singapore. The methods, probabilistic methods and machine
app was available for both Android and iOS learning methods.
platforms. Users who had been recruited from HITS Rules were the first approach category to
were eligible for analyses if they provided at least impute transportation modes [11], [12]. They are
two-week data with at least five-day data validated. sufficient for resolving clear and specific cases but
Finally, 793 out of 1541 recruited participants insufficient for dealing with the ambiguities,
completed the survey with 7856 validated days. between car and bus for example. Probabilistic
Personal information of users was caught from data solutions are more powerful by means of covering
of HITS. and addressing flexibly similarities of behaviors of
In Shanghai (China) [9], an app was created modes. Rules can be considered as specific cases of
to collect data of users over 18 years old. Because it probability-based methods. The typical studies of
was an independent survey, recruitment was this group are based on fuzzy logic theory [13], [14]
undertaken on the internet along with called relatives along with multinomial logistic regression [15]. The
of researchers. Differing from FMS in Singapore, most powerful classifiers are machine learning
users were required to answer questions about their methods like decision tree, random forest, Bayesian
socio-demographics (e.g. age, gender, education, network, support vector machine and
monthly income, frequently visited grocery stores). (convolutional) neural network [16], [17].
The app did not have function allowing users to Compared with deterministic and probabilistic
validate data on their smartphones. Computer- counterparts, machine learning algorithms have
assisted telephone interview were employed to generated very high accuracy but requiring great
gather ground truth. amount of input data together with ground truth to
As an experiment of the National Household train models. Another drawback is lack of
Travel Survey in New Zealand, the Smartphone- interpretability due to every computation taking
Based Individual Travel Survey System (SITSS) place in the black box.
developed an app named ATLAS II for iPhone only In the list of mode, walk, bike, car, bus and
[8]. Similar to FMS, participants in SITSS could train/metro are frequently occurred. The reason is
check their daily itineraries before uploading the geographical venues of previous researches are
labelled trips via Wi-Fi to the SITSS servers. in developed countries. Some studies [10], [18], [19]
Interestingly, users could refuse to give label of a trip focus on classify between non-motorized and
but not delete it. The removal was valid for a whole motorized modes.
day solely. As for variables for classifiers, nearly
th
The app called ITINERUM [10] was maximum instantaneous speed (i.e. 95 percentile),
introduced by Concordia University in Montreal, average and median speed are heavily used along
Canada. Dissimilar to the aforesaid apps, with acceleration profiles. In addition, some features
ITINERUM operates at very high level of silence. are used to enhance the complex models. For
To specific, respondents were not asked to check and example, heading change rate [20], jerk [17], low
validate data every day. This is a key to alleviate rate speed [9] are deployed. To gain a high precision
burden on participants. Users of the app indicated in and recall of bus, GIS data are necessary [12].
[10] were students and colleagues of the authors in
Concordia University. 3. Mobility survey experiment in Hanoi
Based on the review, two main points can be 3.1 Recruitment and data collection
seen in the previous surveys employing smartphones At the end of 2018, Economic and Social
are (1) challenges to recruitment, (2) burden due to Dynamics of Transport Laboratory (DEST) under
1
gathering ground truth. IFSTTAR, France decided to carry out an
2.2 Mode detection methods experiment on observing travel by smartphone in
Hanoi, Vietnam. By a contract with

1
https://www.dest.ifsttar.fr/linstitut/ame/laboratoires/dest-ifsttar/
2 https://www.trivector.se/

37