| dc.description.abstract | Highways are the most frequently used means of transportation in today’s world and the leading source of travel mishaps. Crashes or incidents on highways—both primary and secondary—constrain highway capacity, threaten passenger safety, and increase travel time, resulting in delays and wasted traffic management resources. This thesis aims to expand field knowledge about the detection of secondary incidents by analyzing primary incidents and their spatiotemporal influence on traffic.
Analytical and statistical methods including logit, probit, and artificial neural network models were designed for automating incident classification by processing vehicle count, weather conditions, and traffic flow, among other parameters. The logit and probit model showed similar performance with an accuracy of 67% in the former and 66% in the latter and an identical precision of 48%. The contribution of each independent feature was gauged using odds ratio. The artificial neural network (ANN), on the other hand, out-performed the logit and probit model. A simple 3-layer ANN was used for incident classification which showed an accuracy of 91% and a precision of 89%. The improved performance of ANN can be attributed to its ability to learn complex relations.
A novel connection-weight algorithm was then used to determine the importance of the various features on the dependent variable and how they affect the model. Results were encapsulated in a graphical user interface for facilitating data collection and analysis. | en_US |
