Applications of Bayesian methods for the annual traffic census in Hong Kong

Abstract

Hong Kong has been conducting traffic survey every year since 1965. The survey is known as the Annual Traffic Census (ATC) and is carried out by the Transport Department, Hong Kong S.A.R. Government. One of the main objectives of the ATC is to determine the annual average daily traffic (AADT) of road links in Hong Kong. The AADT can be calculated directly for each core station where data are collected regularly throughout the year. For the short period count stations, known as coverage stations, their AADT have to be estimated by using Group Scaling Factors (GSF) derived from core stations of similar traffic patterns. Traditionally, cluster analysis is used in the ATC for grouping core stations into factor groups. However, in many occasions, it is unable to allocate properly some coverage stations to the factor groups. This study makes use of Bayesian methods for solving the stated problem and for providing an alternative to estimate AADT. It is a method to minimize the likelihood of assigning the station to the wrong cluster. A likelihood function describing the sample count is combined with prior estimates of the probabilities that a station belongs to each cluster. By applying the Bayes' theorem, the posterior classification probabilities are produced. The sample station is then assigned to the cluster showing the highest posterior classification probability. A Bayesian estimate of AADT for the sample station can also be developed. A demonstration of the methodology using actual data of the Annual Traffic Census is carried out to evaluate the applicability of the Bayesian methodology in Hong Kong.

Date of Award	15 Jul 2005
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Hing Po LO (Supervisor)