Conclusion and proposition for future works
At this point we are pleased to discover valuable findings while facing some inevitable obstacles. Moreover, we
studied the flows in Paris Region in order to guide us one step ahead to the final goal since the idea was to look for
dominant trends of the flow likely revealing a certain mode of transportation. Nevertheless this paper represents the
early steps of the transportation mode detection research which is still in progress. Basically the application test was
focused more on metro detection. As a conclusion, here the summarized barriers are presented as well as the final
results of the study.
Over two time period of the data, morning and afternoon, the focus remains on the morning flow because of the
imperfection of the afternoon flow compare to the morning one. The data showed that for about half of the origins
and the destinations, there was no flow recorded in the afternoon period.
It hasn’t been any detected flow between two zones with rather a long distance. Either the share of long distance
trips was not significant or the trips for such trajectories are not in an interest in reality.
In general perspective there is rather no specific concentration of the flow around any group of zones. This means
that the data is rather dispersed in the whole region without an interesting source of clue to conclude a consistent
manner of demand. Nevertheless moving from border to the center, the flow has an increasing trend. Lack of the
public transport in suburb, urbanization issues and socio-economic aspects, more working activities, schools and
tourist attractions could cause this phenomenon.
There are some continuous behaviors detected by the highest range of flow. Investigating the Google map, this is
likely caused by the transit routes, mostly commuter trains, linking the airports from the suburb to the city center,
besides some touristic attractions in suburban area.
As we explicitly disclosed earlier, the metro users could be traced rather easy thanks to the telecom network
structure in Paris. Note that some quality tests are needed to be done in order to detect erroneous entry and exit
points. Our metro detection approach was tested on one day mobile phone data and encouraging results have been
obtained.
The approach presented here is deeply under development. The future works go for more advanced assessment
strategies in its perspective. First of all it is planned to apply the metro detection approach on mobile phone data for
a longer period, more than just a day. Secondly, a step validation will be necessary by comparing the separated
metro flows with reference data from RATP group, the public transport operator in Paris. The validation can be
made only for origins of metro flows, since the provided data by RATP group is rather general containing only the
number of entries per metro station aggregated by year. Thirdly, once the validation is done, it will be interesting to
conduct the urban dynamics studies and the travel demand history with respect to the OD flows of metro users and
work on disaggregation of flows using statistical data.
Another perspective is to apply this approach to detect the other modes of transportation such as trains, cars, etc.
Pedestrian are the most challenging flows and hardly detectable using mobile phone data. Clearly, the records are
depended on the antenna locations and pedestrian walks are quite short by its nature, therefore the probability to
walk around the same antenna coverage is rather high. Nevertheless, for some users and certain segments it might be
possible to manage. For example, it should be assumed that before entering in the metro and just after the user gets
out from the metro, most probably he/she would be walking. Hence the segments having walking mean of
transportation can be inserted by adding such hypothetical locations while the question remains as at what level this
separation of pedestrian flow is significantly important.Conclusion and proposition for future works
At this point we are pleased to discover valuable findings while facing some inevitable obstacles. Moreover, we
studied the flows in Paris Region in order to guide us one step ahead to the final goal since the idea was to look for
dominant trends of the flow likely revealing a certain mode of transportation. Nevertheless this paper represents the
early steps of the transportation mode detection research which is still in progress. Basically the application test was
focused more on metro detection. As a conclusion, here the summarized barriers are presented as well as the final
results of the study.
Over two time period of the data, morning and afternoon, the focus remains on the morning flow because of the
imperfection of the afternoon flow compare to the morning one. The data showed that for about half of the origins
and the destinations, there was no flow recorded in the afternoon period.
It hasn’t been any detected flow between two zones with rather a long distance. Either the share of long distance
trips was not significant or the trips for such trajectories are not in an interest in reality.
In general perspective there is rather no specific concentration of the flow around any group of zones. This means
that the data is rather dispersed in the whole region without an interesting source of clue to conclude a consistent
manner of demand. Nevertheless moving from border to the center, the flow has an increasing trend. Lack of the
public transport in suburb, urbanization issues and socio-economic aspects, more working activities, schools and
tourist attractions could cause this phenomenon.
There are some continuous behaviors detected by the highest range of flow. Investigating the Google map, this is
likely caused by the transit routes, mostly commuter trains, linking the airports from the suburb to the city center,
besides some touristic attractions in suburban area.
As we explicitly disclosed earlier, the metro users could be traced rather easy thanks to the telecom network
structure in Paris. Note that some quality tests are needed to be done in order to detect erroneous entry and exit
points. Our metro detection approach was tested on one day mobile phone data and encouraging results have been
obtained.
The approach presented here is deeply under development. The future works go for more advanced assessment
strategies in its perspective. First of all it is planned to apply the metro detection approach on mobile phone data for
a longer period, more than just a day. Secondly, a step validation will be necessary by comparing the separated
metro flows with reference data from RATP group, the public transport operator in Paris. The validation can be
made only for origins of metro flows, since the provided data by RATP group is rather general containing only the
number of entries per metro station aggregated by year. Thirdly, once the validation is done, it will be interesting to
conduct the urban dynamics studies and the travel demand history with respect to the OD flows of metro users and
work on disaggregation of flows using statistical data.
Another perspective is to apply this approach to detect the other modes of transportation such as trains, cars, etc.
Pedestrian are the most challenging flows and hardly detectable using mobile phone data. Clearly, the records are
depended on the antenna locations and pedestrian walks are quite short by its nature, therefore the probability to
walk around the same antenna coverage is rather high. Nevertheless, for some users and certain segments it might be
possible to manage. For example, it should be assumed that before entering in the metro and just after the user gets
out from the metro, most probably he/she would be walking. Hence the segments having walking mean of
transportation can be inserted by adding such hypothetical locations while the question remains as at what level this
separation of pedestrian flow is significantly important.