- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Theses and Dissertations /
- Application of dynamic artificial neural network to...
Open Collections
UBC Theses and Dissertations
UBC Theses and Dissertations
Application of dynamic artificial neural network to road collision prediction modelling Faghihi, Seyed Farhad
Abstract
Many sets of road traffic data are a collection of ordered measures over the course of time. The key characteristic of time series data is the autocorrelation inside the observed points. Time series analysis is a technique to find the patterns inside time series data, while considering the autocorrelation. The existing time series statistical models in the literature suffer from several assumptions such as stationarity, normality, or homoscedasticity. In this research, a two-stage methodology of Discrete Wavelet Transform (DWT) coupled with Dynamic Artificial Neural Network (DANN) was proposed for predicting a given collision time series. The DANN model enabled the mapping of nonlinear, nonstationary relationship between the future values of collision series and its previous values or, moreover, the previous values of other time series. The DWT facilitated the learning process for limited, noisy observations by decomposing a time series into multiple subseries, which represented specific features of data. The relative performance of the proposed methodology was compared to the conventional AutoRegressive Integrated Moving Average (ARIMA) model for road safety studies. The variables used to build and validate the models were monthly time series data recorded from January 2003 to June 2016 in the City of Kelowna, BC. 54,000 DANN models were trained in total to determine the optimum network configurations for the subseries. The findings of this research showed that the combined DWT and DANN methodology outperformed the ARIMA models in terms of several indices. In addition, the proposed methodology was applied to evaluate the safety phases implemented on Springfield Road, Kelowna at various times. The results demonstrated that the methodology was also capable of using for the intervention analysis, where the traditional time series modelling approaches cannot be executed. The comprehensive study of this thesis is expected to have important implications for both theory and practice in transportation engineering and road safety.
Item Metadata
Title |
Application of dynamic artificial neural network to road collision prediction modelling
|
Creator | |
Publisher |
University of British Columbia
|
Date Issued |
2018
|
Description |
Many sets of road traffic data are a collection of ordered measures over the course of time. The key characteristic of time series data is the autocorrelation inside the observed points. Time series analysis is a technique to find the patterns inside time series data, while considering the autocorrelation. The existing time series statistical models in the literature suffer from several assumptions such as stationarity, normality, or homoscedasticity. In this research, a two-stage methodology of Discrete Wavelet Transform (DWT) coupled with Dynamic Artificial Neural Network (DANN) was proposed for predicting a given collision time series. The DANN model enabled the mapping of nonlinear, nonstationary relationship between the future values of collision series and its previous values or, moreover, the previous values of other time series. The DWT facilitated the learning process for limited, noisy observations by decomposing a time series into multiple subseries, which represented specific features of data. The relative performance of the proposed methodology was compared to the conventional AutoRegressive Integrated Moving Average (ARIMA) model for road safety studies. The variables used to build and validate the models were monthly time series data recorded from January 2003 to June 2016 in the City of Kelowna, BC. 54,000 DANN models were trained in total to determine the optimum network configurations for the subseries. The findings of this research showed that the combined DWT and DANN methodology outperformed the ARIMA models in terms of several indices. In addition, the proposed methodology was applied to evaluate the safety phases implemented on Springfield Road, Kelowna at various times. The results demonstrated that the methodology was also capable of using for the intervention analysis, where the traditional time series modelling approaches cannot be executed. The comprehensive study of this thesis is expected to have important implications for both theory and practice in transportation engineering and road safety.
|
Genre | |
Type | |
Language |
eng
|
Date Available |
2018-10-02
|
Provider |
Vancouver : University of British Columbia Library
|
Rights |
Attribution-NonCommercial-NoDerivatives 4.0 International
|
DOI |
10.14288/1.0372348
|
URI | |
Degree | |
Program | |
Affiliation | |
Degree Grantor |
University of British Columbia
|
Graduation Date |
2018-11
|
Campus | |
Scholarly Level |
Graduate
|
Rights URI | |
Aggregated Source Repository |
DSpace
|
Item Media
Item Citations and Data
Rights
Attribution-NonCommercial-NoDerivatives 4.0 International