A model for estimating pedestrian traffic volumes in urban areas : a case of Seoul

다국어 초록 (Multilingual Abstract)

This study aimed to find a good model to estimate the pedestrian traffic volumes of Seoul as a representative case of large urban areas in Korea. In recent years, increasing concerns about global warming have made people realize a necessity of an sustainable development. Following these trends, researchers in the fields of transportation started studying on non-motorized travel modes, especially pedestrian movements. Those studies mainly aimed to analyze pedestrian safety, satisfaction and movements. Most of those studies noted that an survey on the pedestrian volumes needs to be fulfilled in advance. However, a survey on counting the pedestrian volumes spends a lot of cost and time. In order to overcome those restrictions, some recent studies have tried to make an estimating pedestrian model in order to save cost and time, but it has not been effectively done.
This study proposed two methodologies to build a appropriate pedestrian volume model. The first model was a regression model that had been used in many researches. This model assumes that the pedestrian volumes can be explained and estimated by some pertinent variables, for example accessibility to subway or bus, population and employment nearby the street and pedestrian street features like width. The second model was a neural network model that are combined with a regression modeling. This model is suggested, first in this study, for estimating pedestrian traffic volumes. The aim for using the neural network model is to overcome the weakness of regression models used in this study and to build a better model which has a good predictive power.
The regression models in this study seemed to have good model fits. The F-test was significant at a 99.9 percent confidence level. In addition most of the explanatory variables was statistically significant at the 99 percent confidence level and had a logical interpretations. However, the adjusted -values of the models was 0.443∼0.595, so it indicates the regression model are still not appropriate to estimate the pedestrian volumes.
In order to get more predictive power to estimate the pedestrian volumes, this study suggested a neural network model combined with a regression modeling. In fact, the neural network models are well known for their good estimation power in various research fields. However, the neural network models have a critical problem that there is no principle of choosing input variables. Therefore, this study put a regression modeling technic in the neural network model to solve the selecting input variable problems. Consequently, the neural network models showed better results than the regression models from the test results. However, from the validation results, the neural network models could not show good predictive powers. From those results, it can be implied that if we combine the neural network model's high predictive power and the statistical procedures of regression models, we can get better results.
In conclusion, the estimated parameter values of the regression models and the relative importance of the input variables of the neural network models can inform us of the relationship between the pedestrian volumes and pertinent variables. This study could contribute to some other future studies. For example, the results of pedestrian volume estimation can be used in market analysis. The pedestrian volumes and rental cost of building nearby pedestrian street are closely related, but we do not know how much the effects exist. In addition, this study can be used in pedestrian level of service (LOS) analysis which was very hard because the cost of survey on pedestrian volumes are so expansive.

목차 (Table of Contents)

• 1. Introduction 1
• 1.1 Background and purpose 1
• 1.2 Scope of the study 2
• 1.3 Research flow 5
• 2. Literature review 6
• 2.1 Previous pedestrian volume models 6
• 2.2 Neural network theory 11
• 3. Methodology 16
• 3.1 Introduction 16
• 3.2 Pedestrian estimating models 18
• 3.3 Technic to compare models 32
• 4. Model application 33
• 4.1 Regression model 33
• 4.2 Neural network models 44
• 4.3 Test and Validation results 51
• 5. Conclusions 52
• ■ References 56
• Appendix A : Description of data features 59
• Appendix B : Additional regression analysis technic 65
• Appendix C : Additional results of regression models 69
• Appendix D : Established structures of neural network models (NM2) 76