Conference Papers 2013

Seminar
Day 1 (30 Sep 2013), Session 2, Activity Based Modelling, 16:30 - 17:00

Status
Accepted, awaiting documents

Submitted by / Abstract owner
Taha Hossein Rashid

Authors
Taha Hossein Rashidi, University Of New South Wales, Hironobu Hasegawa, Akita National College of Technology

Short abstract
There are several issues involved with these recent travel simulation frameworks that require further investigation before they become fully operational. This study attempts to address these issues while discussing some advanced data mining methods to improve the goodness-of-fit of the models.

Abstract
Introduction and Background
Since the introduction of activity-based models (Kitamura 1988) several applications of them have been developed. At the same time the conventional and popular four step models are being less studied and developed by researchers and practitioners due to their policy sensitivity restrictions (Axhausen and Garling, 1992). Nonetheless, the cost, including time and money, of developing the highly disaggregate activity-based models is considerably higher than developing the aggregate models (Pendyala et al 2010). As a result, especially for small and medium sized cities, the need for a disaggregate travel demand model which is not costly as an activity-based model seems to be essential. Such a disaggregate travel demand model should replace the trip generation, trip distribution and mode choice steps of an aggregate four-step model while it is not as aggregate as four-step models. There has been some recent attempts to bridge this gap between four-step models and activity-based models (Rashidi and Mohammadian 2011, Langeroudi et al. 2012) by developing data transferability models that can be used to simulate household or individual level travel attributes. There are several issues involved with these recent travel simulation frameworks that require further investigation before they become fully operational. This study attempts to address these issues while discussing some advanced data mining methods to improve the goodness-of-fit of the models.
First, it is essential that mode choice, trip purpose, time of day and trip distance models should be jointly modelled as these decisions have a high inter-dependency. Consideration of this complex structure of correlations among these decisions was left for future research in the previous studies (Rashidi and Mohammadian 2011, Langeroudi et al. 2012) and will be addressed in this paper. Second, a destination choice model is an essential element of a complete set of trip making decisions which was left unaddressed in the previously studies and will be discussed in the study framework of this paper. Finally, decision tree method was the methodology that was used in the previous papers and will be substituted with a more advanced method of random forest (Ho, 2002; Prinzie and Van Den Poel, 2008) in the current paper.

Data and Methodology
The main data sources used in the study are the household travel surveys collected for major Australian cities of Sydney, Melbourne and Brisbane (DOT 2009, DOT 2011 and Transport NSW 2010). Other supporting data sources like land-use data will also be used to compliment the data of household travel surveys.
Data mining approaches have now been increasingly applied in many academic fields for prediction models. Data mining is defined as an approach which explores large datasets to find consistent and systematic interdependencies among the variables. Among data mining approaches are rule based methods which construct the relationship between the dependent variable and the independent ones using certain rules. Such rules cause the data to be categorized in a number of clusters in a way that data in each cluster share homogenous attributes. Rashidi and Mohammadian (2011) used the exhaustive CHAID method to find the optimal decision tree for the several travel attributes. Following that study, Langeroudi et al. (2012) used the same method to find the best decision trees for trip generation rates for several trip purposes. In the current paper a sophisticated method which is drawn from the concept of the decision tree, namely random forest (Breiman 2001) is used to find the best fit to the data. To the best of the authors’ knowledge, this is the first application of these methods in travel demand modelling. It was found that random forest can provide 88% accuracy when it is applied on the Melbourne data for the abovementioned four travel attributes when they are jointly modelled which is a significant achievement.

References
Axhausen, K., Garling, T.: Activity-based approaches to travel analysis: conceptual frameworks, models, and research problems. Transp. Rev. 12(4), 323–341 (1992)

Breiman, L. (2001) Random Forests, Machine Learning, 45 (1): 5–32

DOT – Department of Transport (2009) Victorian Integrated Survey of Travel and Activity 2007, Melbourne: DOT

DOT – Department of Transport (2011) http://www.transport.vic.gov.au/vista , VISTA website Department of Transport, last accessed 10/22/2011

Hastie, T., Tibshirani, R., Friedman, J. H. (2009) Boosting and Additive Trees, The Elements of Statistical Learning (2nd ed.). New York: Springer: 337–384

Ho, Tin Kam (2002) A Data Complexity Analysis of Comparative Advantages of Decision Forest Constructors, Pattern Analysis and Applications, 5:102-112

Kitamura R. (1988) An evaluation of activity-based travel analysis, Transportation, 15(1): 9-34

Langeroudi M. F., T. H. Rashidi and A. Mohammadian (2012) Individual Trip Rate Transferability Analysis: A Decision Tree Approach, 13th International Conference on Travel Behaviour Research, Toronto 15-20, July 2012

Pendyala R., C. Chiu Y., P. Waddell, M. Hickman, K. Konduri and B. Sana, The design of an integrated model of urban continuum-location choices activity travel behavior and dynamic traffic patterns, 12th WCTR, July 11-15, 2010 – Lisbon, Portugal

Prinzie, A. and Van Den Poel, D. (2008) Random Forests for multiclass classification: Random MultiNomial Logit, Expert Systems with Applications, 34 (3): 1721-1732

Rashidi, T.H. , Mohammadian, A.(2011), Household travel attributes transferability analysis: application of a hierarchical rule based approach, Transportation, 38:697–714

Transport NSW, Transport Data Centre (2010) 2008/09 Household Travel Survey Summary Report 2010 Release; REPORT 2010/01, JUNE 2010; Sydney: Transport NSW