Detecting Activity Types and Trip Purposes from Passive GPS Data: A Data Mining Approach - Space-Time Integration in Geography and GIScience - page 212

to be more active than females in recreation activities such as sports while women
prefer to spend more time on maintenance activities such as shopping and child-
care. Buliung ( 2005 ) reveal that the elder and married people with young children
are less likely engaged in out-of-home recreational activities than others.
The above discussions show that trip purposes can be identified by the character-
istics of the trip and the socio-economic characteristics of the person who makes the
trip. Thus, the detection of trip purpose can be treated as a classification problem that
match trip purposes with the corresponding trip characteristics (e.g. trip duration
and timing) as well as individuals' socioeconomic characteristics. In this regard, the
genetic algorithm is applied to establish the classification models. In the following,
we shall explain how the GA may be used to establish the classification models.
The classification models can be considered as a series of heuristic rules in the form
of “IF, THEN”. The “IF” part (i.e., the rule antecedent) contains a combination of
conditions on attributes and values, whereas the rule consequent (the “THEN” part)
contains goal attribute (Noda et al. 1999 ). In this case, land use type of trip ends,
trip duration and timing as well as individuals' socioeconomic characteristics may
be used as the attributes in the “IF” part, whereas categories of trip purpose are
considered as the goal attribute in the “THEN” part.

13.3.2.1 Encoding Chromosome

The classification rules of GA are represented by chromosomes. There are two
approaches, namely Michigan and Pittsburgh approaches that are commonly used
to encode chromosomes. By the Michigan approach, each classification rule is
represented by one chromosome, whereas by the Pittsburgh approach, a set of
classification rules are encoded in a single chromosome (Freitas 2002 ). For the trip
purpose detection problem, if the Michigan approach is applied, one chromosome
is developed for each trip purpose separately. For example, the chromosome for the
trip purpose 'work' may looks like:

IF (7:30 < start time < 9:00 C 17:00 < end time < 20:00 C 4 hours < duration <

10 hours ::: :::)THEN“ work ”.

On the other hand, if the Pittsburgh approach is used, a single chromosome will
set the rules for detecting all the trip purposes. Such a chromosome may looks like:

IF (18:00 < start time < 23:00 C 7:00 < end time < 9:00 C 7 hours < duration <

12 hours ::: :::)THEN“ home ”; IF (7:30 < start time < 9:00 C 17:00 < end
time < 20:00 C 4 hours < duration < 10 hours ::: :::)THEN“ work ”; IF
(17:30 < start time < 20:00 C 19:00 < end time < 20:00 C 1 hour < duration <
5 hours ::: :::)THEN“ recreation ” :::).

While the Pittsburgh approach has the advantages of generating all classification
rules at once and allowing interactions between classification rules and the eval-
uation of classification rules as a whole, its disadvantages are: the chromosomes