Detecting Activity Types and Trip Purposes from Passive GPS Data: A Data Mining Approach - Space-Time Integration in Geography and GIScience - page 224

The classification models developed can now be applied to detect trip purposes
for given data on land use, starting time, duration, age and day of the week.
There might be different ways to apply the models for detection. One way is to
assign the purpose of a classification model that has the closest distance to the
given data. The Euclidean distance can be used to measure the differences between
the data to be classified and different classification models. Thus, if the purpose
of the classification model falls within the possible purposes of a trip end, we
can then calculate the distance by adding the distances from the given data on
starting time and duration to those values specified in the classification models of
possible purposes. The smaller the gap it is with one classification model, the higher
possibility it is classified as the trip purpose of the model. The distance concerning
the starting time can be acquired directly by the point-to-point comparison. As for
that concerning duration, if the given data lies within the minimum and maximum
value of duration in the classification model, the distance for duration would be zero.
Otherwise, distances will be calculated respectively between the given duration and
the two boundary values in the model. The shorter distance is considered to be the
distance from the duration boundaries of the model.

13.5.4 Internal and External Validity

To assess the applicability of a model or algorithm, it is important to evaluate
its validity. Cook and Campbell ( 1979 ) elaborate the types of validity and define
that internal validity addresses whether or not an observed covariation could be
considered a causal relationship. In contrast to internal validity, external validity
evaluate whether or not an observed causal relationship could be generalized to dif-
ferent measures, persons, settings, and times. For relatively small data sets, internal
validation of models may not be sufficient and indicative for their performance in
other studies. External validation may therefore be necessary before implementing
these models in practice (Bleeker et al. 2003 ). To evaluate the validity of the
algorithm developed in this study, we can compare the trip purposes detected by
the algorithm to the real trip purposes reported by respondents. The 1-min data set
(i.e., a total of 549 activities) is used to develop the classification models. These
models are then applied to detect trip purposes for the 1-min dataset to assess
internal validity and for the 3-min dataset (about 310 activities) to evaluate external
validity. The results on internal validity are presented in Table 13.2 and that on
external validity in Table 13.3 .
Tab le 13.2 shows the cross-tabulation of the trip purposes reported by respon-
dents and that detected by the models. The accuracy rate for each category is also
reported. As one may tell from the table, the internal validity of the model is in
general very high. The accuracy rates for three of the seven categories are higher
than 85 %. As expected, home and school have the highest accuracy rates. The
accuracy rates for shopping and recreation are also rather high. On the other hand,