Now we discuss Figure 10 from the perspective of transportation
network data availability. Depending on its availability, the
transportation network data can be categorized into three levels.
The most widely available data is network topology data such as
rail line routes. Figure 10 shows that this data is also most useful
among transportation network features for mode detection. This is a
good property of our approach. It means that our approach can be
deployed to many regions in the world and is likely to achieve good
performance there. The less widely available data is bus stop
locations. Figure 10 shows that this data is least useful among the
top ranked features. This means that our approach would not lose
too much performance in the regions where bus stop information is
unavailable. The least available data is real-time bus locations,
which is a very predictive feature (i.e., average bus closeness)
according to Figure 10. Thus our approach will not be able to
utilize this predictive feature in many regions of the world. This is a
limitation of our approach.