Geography Reference
In-Depth Information
To calibrate the algorithm, we need to decide on the size of the population,
maximum number of generations, and the probabilities of crossover and mutation. It
should be noted that similar to other machine learning methods, parameters in GA
such as population size, maximum generation, as well as crossover and mutation
probability (Pc and Pm) need to be adjusted and improved continuously during
the evolutionary process. Obviously, if the population size is set unreasonably, it's
difficult to find out the optimum solution or the convergence time would be greatly
extended. In this study, the population size is set to be between 20 and 30. GA is
sensitive to the crossover and mutation rates, a crossover rate that is too high may
lead to premature convergence of the genetic algorithm, in contrast, it may slow
down the progress of evolution (Kaplan and Hegarty
2005
). Similarly, a very small
mutation rate may lead to genetic drift while a mutation rate that is too high may
result in the loss of some good solutions. Lots of experiments in the previous studies
indicate that Pc value should range from 0.5 to 0.9 while Pm value should be set
between 0.05 and 0.2 ( Sen and Öztopal
2001
;Yangetal.
2006
).
13.5.3
Results
Figure
13.1
shows part of the algorithm's output file, which includes the input
parameters in the top left box as well as the first three and last five generations
of the chromosome evolution. As can be seen in the bottom right box, after 80
generations' evolution in this run, we got the best classification model with the
fitness value of 0.7347 for the purpose of “Recreation”: 0.7793, 0.0127, 0.1432,
and 5.0000, which indicates 18:42 as start time, 18 min minimum duration, 3 h and
26 min maximum duration and trip purpose 'recreation' respectively. The whole
sentence can be interpreted as: on condition that “5” is included as one of the
possible purposes, if activity starts after 18:42 while its time duration falls into the
intervalbetween18minand3hand26min,itisregardedtobethebestmodelto
identify trip purpose as “Recreation”.
After running dozens of times for each trip purpose category, best classification
models having the highest fitness value for each category can be identified.
Tab le
13.1
lists the best classification model for all categories of trip purposes
specifically for each of the four cases identified earlier. Information on which run of
the algorithm and which generation the best model was obtained is also given in the
table. No classification models are given for the cases in which the trip purposes are
not reported by respondents.
As shown in Table
13.1
, classification models for activity types/trip purposes
such as “Home”, “Work” and “School” have rather high value of fitness, higher than
that for “shopping”, “Recreation” and “Personal affairs”. This may be due to fact
that the land use data for home, work and school are more complete and updated
than that for other trip purposes. Another possible reason is that activity types of
shopping, recreation and personal affairs are more likely performed at places with
mixed land use than other types. It is usually more difficult to identify trip purposes
Search WWH ::
Custom Search