Page 144 - Proceeding of Atrans Young Researcher's Forum 2019_Neat

“Transportation for A Better Life:
Smart Mobility for Now and Then”

23 August 2019, Bangkok, Thailand

seen that in the case of RMSE, both SVM and EDT deviations StD of RMSE, MAE and accuracy
Bagged possessed very close probability density distribution using EDT Bagged is slightly lower than
distribution lines, which reached a peak at RMSE = those using SVM (StD of RMSE = 0.0714 and
0.9 for SVM and RMSE = 0.88 for EDT Bagged. In 0.0716, StD of MAE = 0.0434 and 0.0442, StD of
terms of MAE criterion, SVM algorithm reached a accuracy = 0.0219 and 0.0223 using EDT Bagged
more important value compared to EDT Bagged and SVM respectively). The higher values of
(e.g. MAE=0.38 for SVM and MAE = 0.33 for EDT standard deviation, the low degree of robustness the
Bagged at max frequency). In terms of accuracy, method exhibits. With a lower value of RMSE, MAE
EDT Bagged reached a higher peak at accuracy = and higher value of Accuracy, EDT Bagged model
0.8, whereas that of SVM was at accuracy = 0.76- is more robust than SVM in prediction the travel
0.77. Regarding the level of fluctuation, standard decisions of transport users.

Figure 8. Graphs of the histogram for 1000 Monte Carlo simulations in case of: (a) RMSE; (b) MAE;
(c) Accuracy;

transport users. These methods allow the modal
Based on the analysis of 1000 Monte Carlo
simulations, with respect to the mean values, the decision-making process to be examined and
probability density function and the fluctuations of understood in great detail. However, EDT
RMSE, MAE, Accuracy criteria, it can be easily Bagged yielded a slightly better result than
deduced that EDT Bagged outperforms SVM. SVM. From overall statistical analysis (mean,
max frequency and standard deviation) of
4. Conclusions RMSE, MAE and accuracy based on Monte
Carlo simulations, EDT Bagged model is more
In this study, AI approaches namely robust than SVM in prediction the travel
EDT Bagged and SVM have been proposed for decisions of transport users.
prediction of travel decisions of transport users
in Hanoi, Vietnam. A comprehensive For conclusions, the results of this
understanding of the characteristics of travel study would be useful for transport planners to
demand and the possible impacts of income, potentially apply AI methods in transport
trip mode, trip purpose, trip length, trip cost, demand modeling. Such interdisciplinary
and parking charge on travel decisions of mobility research shows advances in predicting
transport users was analyzed to construct the travel decisions. This study is also beneficial for
database. From the obtained results of transport authorities to formulate effective
prediction capability, it revealed that both SVM transport management strategies for mode-
and EDT Bagged algorithms are potential shifting encouragement to achieve urban
candidates for predicting the travel decisions of transport sustainability.

119