Therefore, we learn the Bayesian network
structure and estimate the corresponding set of conditional
probability tables from the GPS track data collected in a
smartphone-based travel survey. Subsequently, the travel modes
are automated on the basis of the resulting Bayesian network.
We additionally improve the classifier using the confusion matrix
derived from the original classifier by introducing two targeted
variables. It is expected that a promising accuracy can be achieved
for mode detection and that our study of GPS travel surveys will
provide an opportunity to supplement traditional travel surveys.