Database Reference
In-Depth Information
in comparison with female oriented rules. The smaller number and attraction of
male-oriented stores will certainly be one of the main reasons as to why we see
this female bias. Clearly, more research is needed in order to mine for interesting
patterns that instead of stating the obvious should provide interesting and new
knowledge.
14.4 Conclusions
In this chapter, we have demonstrated the merits of Bluetooth tracking as an
innovative, inexpensive, unobtrusive, and flexible methodology for measuring
human mobility in a variety of contexts and environments. At mass events it can
aid crowd managers by delivering quantitative data on crowd sizes and flows,
and in retail environments it can extract marketing intelligence or other organi-
zational intelligence through methods ranging from visual data exploration to
data mining techniques such as association rule learning.
However, the unobtrusive nature of the tracking process resulting in large
sample sizes automatically also constitutes a methodological issue: the possi-
bility of biased results by oversampling certain segments of the total population
of individuals. Adolescents with a higher education might indeed carry more
Bluetooth-enabled devices than elderly people, and young children will probably
never be detected. The potential difference in Bluetooth usage among different
audiences might significantly influence generated insights. Accordingly, more
research is needed into the use of discoverable Bluetooth-enabled devices by
different population segments in order for Bluetooth tracking to evolve into
a technology delivering accurate and reliable information to policy makers,
crowd managers, and marketing researchers. The penetration rates we found in
our experiments ranged from around 11% for a general audience to 35% for a
professional fair visitor profile. In the end, a more systematic way of calculat-
ing the percentage of the population being tracked will be necessary for more
reliable extrapolations in the future. Additionally, the possible influence of time
and space on the detection ratio needs to be investigated.
The tentative association rule analysis with the shopping mall data only
shows a very small selection of data mining possibilities with Bluetooth tracking
data. Specifically for association rules, it soon became clear that there is a
need for methods that can filter out more interesting rules from a larger set
of less interesting rules. Intelligent visualization and/or pruning of association
rules instead of solely listing them will certainly aid in this process. Besides
association rule discovery, other data mining methods such as those described
earlier in Chapter
6
can also generate valuable knowledge from this type of sparse
movement data. They might need further modifications, however, to handle the
spatio-temporal complexity of Bluetooth tracking data.
