Geography Reference
In-Depth Information
13.6
Discussions and Conclusions
Over the past few years, although GPS-enabled methods have attracted increasing
attention in the collection of individuals' activity-travel behavior data, the identifi-
cation of activity type and trip purpose is still considered as a bottleneck and keeps
passive GPS from large-scale applications in behavior studies. It was against this
background that this study was initiated and conducted. We developed a new method
based on a genetic algorithm for the detection of trip purpose. We established a set of
classification models through self-learning from data on land use type of trip ends,
trip duration and timing, socioeconomic characteristics of respondents, and other
relevant information. To test the applicability and validity of the proposed method,
we conducted a field study and collected data in 2011. The data were used to test
the applicability and validity of the method. The evaluation results indicated that
more than 80 % of the major trip purposes or activity types have been successfully
detected based on the proposed method. Both the internal and external validity
tests have demonstrated satisfactory ability of this proposed method for identifying
activity type and trip purpose. Most importantly, this learning method has great
potentials to achieve better performance when more data are involved.
This study therefore contributes to the improvement of trip purpose detection by
developing a self-learning method with less artificial interference and establishing
classification models separately according to different characteristics of respondent
and situation.
It should be noted that as an exploratory attempt, this study unavoidably has
limitations. In particular, accuracy of trip purpose detection varied considerably
according to different categories, which was mainly caused by the lack of data for
some land-use types. The success rate of trip purpose detection therefore depends
heavily on the accuracy and completeness of GIS data, especially for those land-use
features. Furthermore, although the overall accuracy already looks promising in this
pilot test, this approach to identify trip purpose from GPS data is at the experimental
stage which needs to be improved in both algorithm design and sample method.
Therefore, several issues may be identified as concerns for future research.
In the short term, participants with wider range of identities and age groups
should be recruited so that more data in both quantity and diversity can be
collected for training and testing. Input data of the algorithm including the variables
and personal information should be more diversified according to different trip
purpose. The design and parameters of GA also need to be improved, in particular,
respondents' socio-economic characteristic such as sex, age and occupation should
be considered as variables within the algorithm so as to participate in the evolution
process rather than just establish different classification models for each case as
employed in this study. Although the process of introducing more variables into
GA would be challenging, to effectively minimize arbitrariness and subjectivity as
typically revealed in previous studies and achieve better classification models, the
benefit of doing so is worth the efforts. We believe that the development of advanced
Search WWH ::




Custom Search